feat：新增 A_Memorix 记忆插件

引入 A_Memorix 插件（v2.0.0）——一个轻量级的长期记忆提供器。新增插件清单（manifest）和入口（AMemorixPlugin），并提供完整的核心能力：嵌入（基于哈希的 EmbeddingAPIAdapter、EmbeddingManager、预设）、检索（双路径检索器、PageRank、图关系召回、BM25 稀疏索引、阈值与融合配置）、存储与元数据层，以及大量实用工具和迁移/转换脚本。同时更新 .gitignore 以允许 /plugins/A_memorix。该变更为在宿主应用中实现统一的记忆摄取、检索、分析与维护奠定了基础。

.gitignore

@@ -339,6 +339,7 @@ run_pet.bat
 
 /plugins/*
 !/plugins
+!/plugins/A_memorix
 !/plugins/hello_world_plugin
 !/plugins/emoji_manage_plugin
 !/plugins/take_picture_plugin

plugins/A_memorix/__init__.py

@@ -0,0 +1,12 @@
+"""
+A_Memorix - 轻量级知识库插件
+
+完全独立的记忆增强系统，优化低资源环境下的知识存储与检索。
+"""
+
+__version__ = "2.0.0"
+__author__ = "A_Dawn"
+
+from .plugin import AMemorixPlugin
+
+__all__ = ["AMemorixPlugin"]

plugins/A_memorix/_manifest.json

@@ -0,0 +1,62 @@
+{
+  "manifest_version": 1,
+  "name": "A_Memorix",
+  "version": "2.0.0",
+  "description": "MaiBot SDK 长期记忆插件，负责统一检索、写入、画像与记忆维护。",
+  "author": {
+    "name": "A_Dawn"
+  },
+  "license": "AGPL-3.0",
+  "repository_url": "https://github.com/A-Dawn/A_memorix/",
+  "host_application": {
+    "min_version": "1.0.0"
+  },
+  "keywords": [
+    "memory",
+    "knowledge",
+    "retrieval",
+    "profile",
+    "episode"
+  ],
+  "categories": [
+    "Memory",
+    "Data"
+  ],
+  "plugin_info": {
+    "is_built_in": false,
+    "plugin_type": "memory_provider",
+    "components": [
+      {
+        "type": "tool",
+        "name": "search_memory",
+        "description": "搜索长期记忆"
+      },
+      {
+        "type": "tool",
+        "name": "ingest_summary",
+        "description": "写入聊天摘要"
+      },
+      {
+        "type": "tool",
+        "name": "ingest_text",
+        "description": "写入普通长期记忆文本"
+      },
+      {
+        "type": "tool",
+        "name": "get_person_profile",
+        "description": "查询人物画像"
+      },
+      {
+        "type": "tool",
+        "name": "maintain_memory",
+        "description": "维护记忆关系"
+      },
+      {
+        "type": "tool",
+        "name": "memory_stats",
+        "description": "查询记忆统计"
+      }
+    ]
+  },
+  "capabilities": []
+}

plugins/A_memorix/core/__init__.py

@@ -0,0 +1,84 @@
+"""核心模块 - 存储、嵌入、检索引擎"""
+
+# 存储模块（已实现）
+from .storage import (
+    VectorStore, 
+    GraphStore, 
+    MetadataStore,
+    ImportStrategy,
+    KnowledgeType,
+    parse_import_strategy,
+    resolve_stored_knowledge_type,
+    detect_knowledge_type,
+    select_import_strategy,
+    should_extract_relations,
+    get_type_display_name,
+)
+
+# 嵌入模块（使用主程序 API）
+from .embedding import (
+    EmbeddingAPIAdapter,
+    create_embedding_api_adapter,
+)
+
+# 检索模块（已实现）
+from .retrieval import (
+    DualPathRetriever,
+    RetrievalStrategy,
+    RetrievalResult,
+    DualPathRetrieverConfig,
+    TemporalQueryOptions,
+    FusionConfig,
+    GraphRelationRecallConfig,
+    RelationIntentConfig,
+    PersonalizedPageRank,
+    PageRankConfig,
+    create_ppr_from_graph,
+    DynamicThresholdFilter,
+    ThresholdMethod,
+    ThresholdConfig,
+    SparseBM25Index,
+    SparseBM25Config,
+)
+from .utils import (
+    RelationWriteService,
+    RelationWriteResult,
+)
+
+__all__ = [
+    # Storage
+    "VectorStore",
+    "GraphStore",
+    "MetadataStore",
+    "ImportStrategy",
+    "KnowledgeType",
+    "parse_import_strategy",
+    "resolve_stored_knowledge_type",
+    "detect_knowledge_type",
+    "select_import_strategy",
+    "should_extract_relations",
+    "get_type_display_name",
+    # Embedding
+    "EmbeddingAPIAdapter",
+    "create_embedding_api_adapter",
+    # Retrieval
+    "DualPathRetriever",
+    "RetrievalStrategy",
+    "RetrievalResult",
+    "DualPathRetrieverConfig",
+    "TemporalQueryOptions",
+    "FusionConfig",
+    "GraphRelationRecallConfig",
+    "RelationIntentConfig",
+    "PersonalizedPageRank",
+    "PageRankConfig",
+    "create_ppr_from_graph",
+    "DynamicThresholdFilter",
+    "ThresholdMethod",
+    "ThresholdConfig",
+    "SparseBM25Index",
+    "SparseBM25Config",
+    "RelationWriteService",
+    "RelationWriteResult",
+]
+

plugins/A_memorix/core/embedding/__init__.py

@@ -0,0 +1,18 @@
+"""嵌入模块 - 向量生成与量化"""
+
+# 新的 API 适配器（主程序嵌入 API）
+from .api_adapter import (
+    EmbeddingAPIAdapter,
+    create_embedding_api_adapter,
+)
+
+from ..utils.quantization import QuantizationType
+
+__all__ = [
+    # 新的 API 适配器（推荐使用）
+    "EmbeddingAPIAdapter",
+    "create_embedding_api_adapter",
+    # 量化
+    "QuantizationType",
+]
+

plugins/A_memorix/core/embedding/api_adapter.py

@@ -0,0 +1,174 @@
+"""
+Hash-based embedding adapter used by the SDK runtime.
+
+The plugin runtime cannot import MaiBot host embedding internals from ``src.chat``
+or ``src.llm_models``. This adapter keeps A_Memorix self-contained and stable in
+Runner by generating deterministic dense vectors locally.
+"""
+
+from __future__ import annotations
+
+import hashlib
+import re
+import time
+from typing import List, Optional, Union
+
+import numpy as np
+
+from src.common.logger import get_logger
+
+
+logger = get_logger("A_Memorix.EmbeddingAPIAdapter")
+
+_TOKEN_PATTERN = re.compile(r"[A-Za-z0-9_\u4e00-\u9fff]{1,}")
+
+
+class EmbeddingAPIAdapter:
+    """Deterministic local embedding adapter."""
+
+    def __init__(
+        self,
+        batch_size: int = 32,
+        max_concurrent: int = 5,
+        default_dimension: int = 256,
+        enable_cache: bool = False,
+        model_name: str = "hash-v1",
+        retry_config: Optional[dict] = None,
+    ) -> None:
+        self.batch_size = max(1, int(batch_size))
+        self.max_concurrent = max(1, int(max_concurrent))
+        self.default_dimension = max(32, int(default_dimension))
+        self.enable_cache = bool(enable_cache)
+        self.model_name = str(model_name or "hash-v1")
+        self.retry_config = retry_config or {}
+
+        self._dimension: Optional[int] = None
+        self._dimension_detected = False
+        self._total_encoded = 0
+        self._total_errors = 0
+        self._total_time = 0.0
+
+        logger.info(
+            "EmbeddingAPIAdapter 初始化: model=%s, batch_size=%s, dimension=%s",
+            self.model_name,
+            self.batch_size,
+            self.default_dimension,
+        )
+
+    async def _detect_dimension(self) -> int:
+        if self._dimension_detected and self._dimension is not None:
+            return self._dimension
+        self._dimension = self.default_dimension
+        self._dimension_detected = True
+        return self._dimension
+
+    @staticmethod
+    def _tokenize(text: str) -> List[str]:
+        clean = str(text or "").strip().lower()
+        if not clean:
+            return []
+        return _TOKEN_PATTERN.findall(clean)
+
+    @staticmethod
+    def _feature_weight(token: str) -> float:
+        digest = hashlib.sha256(token.encode("utf-8")).digest()
+        return 1.0 + (digest[10] / 255.0) * 0.5
+
+    def _encode_single(self, text: str, dimension: int) -> np.ndarray:
+        vector = np.zeros(dimension, dtype=np.float32)
+        content = str(text or "").strip()
+        tokens = self._tokenize(content)
+        if not tokens and content:
+            tokens = [content.lower()]
+        if not tokens:
+            vector[0] = 1.0
+            return vector
+
+        for token in tokens:
+            digest = hashlib.sha256(token.encode("utf-8")).digest()
+            bucket = int.from_bytes(digest[:8], byteorder="big", signed=False) % dimension
+            sign = 1.0 if digest[8] % 2 == 0 else -1.0
+            vector[bucket] += sign * self._feature_weight(token)
+
+            second_bucket = int.from_bytes(digest[12:20], byteorder="big", signed=False) % dimension
+            if second_bucket != bucket:
+                vector[second_bucket] += (sign * 0.35)
+
+        norm = float(np.linalg.norm(vector))
+        if norm > 1e-8:
+            vector /= norm
+        else:
+            vector[0] = 1.0
+        return vector
+
+    async def encode(
+        self,
+        texts: Union[str, List[str]],
+        batch_size: Optional[int] = None,
+        show_progress: bool = False,
+        normalize: bool = True,
+        dimensions: Optional[int] = None,
+    ) -> np.ndarray:
+        _ = batch_size
+        _ = show_progress
+        _ = normalize
+
+        started_at = time.time()
+        target_dimension = max(32, int(dimensions or await self._detect_dimension()))
+
+        if isinstance(texts, str):
+            single_input = True
+            normalized_texts = [texts]
+        else:
+            single_input = False
+            normalized_texts = list(texts or [])
+
+        if not normalized_texts:
+            empty = np.zeros((0, target_dimension), dtype=np.float32)
+            return empty[0] if single_input else empty
+
+        try:
+            matrix = np.vstack([self._encode_single(item, target_dimension) for item in normalized_texts])
+            self._total_encoded += len(normalized_texts)
+            self._total_time += time.time() - started_at
+        except Exception:
+            self._total_errors += 1
+            raise
+
+        return matrix[0] if single_input else matrix
+
+    def get_statistics(self) -> dict:
+        avg_time = self._total_time / self._total_encoded if self._total_encoded else 0.0
+        return {
+            "model_name": self.model_name,
+            "dimension": self._dimension or self.default_dimension,
+            "total_encoded": self._total_encoded,
+            "total_errors": self._total_errors,
+            "total_time": self._total_time,
+            "avg_time_per_text": avg_time,
+        }
+
+    def __repr__(self) -> str:
+        return (
+            f"EmbeddingAPIAdapter(model_name={self.model_name}, "
+            f"dimension={self._dimension or self.default_dimension}, "
+            f"total_encoded={self._total_encoded})"
+        )
+
+
+def create_embedding_api_adapter(
+    batch_size: int = 32,
+    max_concurrent: int = 5,
+    default_dimension: int = 256,
+    enable_cache: bool = False,
+    model_name: str = "hash-v1",
+    retry_config: Optional[dict] = None,
+) -> EmbeddingAPIAdapter:
+    return EmbeddingAPIAdapter(
+        batch_size=batch_size,
+        max_concurrent=max_concurrent,
+        default_dimension=default_dimension,
+        enable_cache=enable_cache,
+        model_name=model_name,
+        retry_config=retry_config,
+    )

plugins/A_memorix/core/embedding/manager.py

@@ -0,0 +1,510 @@
+"""
+嵌入管理器
+
+负责嵌入模型的加载、缓存和批量生成。
+"""
+
+import hashlib
+import pickle
+import threading
+from concurrent.futures import ThreadPoolExecutor, as_completed
+from pathlib import Path
+from typing import Optional, Union, List, Dict, Any, Tuple
+
+import numpy as np
+
+try:
+    from sentence_transformers import SentenceTransformer
+    HAS_SENTENCE_TRANSFORMERS = True
+except ImportError:
+    HAS_SENTENCE_TRANSFORMERS = False
+
+from src.common.logger import get_logger
+from .presets import (
+    EmbeddingModelConfig,
+    get_custom_config,
+    validate_config_compatibility,
+    are_models_compatible,
+)
+from ..utils.quantization import QuantizationType
+
+logger = get_logger("A_Memorix.EmbeddingManager")
+
+
+class EmbeddingManager:
+    """
+    嵌入管理器
+
+    功能：
+    - 模型加载与缓存
+    - 批量生成嵌入
+    - 多线程/多进程支持
+    - 模型一致性检查
+    - 智能分批
+
+    参数：
+        config: 模型配置
+        cache_dir: 缓存目录
+        enable_cache: 是否启用缓存
+        num_workers: 工作线程数
+    """
+
+    def __init__(
+        self,
+        config: EmbeddingModelConfig,
+        cache_dir: Optional[Union[str, Path]] = None,
+        enable_cache: bool = True,
+        num_workers: int = 1,
+    ):
+        """
+        初始化嵌入管理器
+
+        Args:
+            config: 模型配置
+            cache_dir: 缓存目录
+            enable_cache: 是否启用缓存
+            num_workers: 工作线程数
+        """
+        if not HAS_SENTENCE_TRANSFORMERS:
+            raise ImportError(
+                "sentence-transformers 未安装，请安装: "
+                "pip install sentence-transformers"
+            )
+
+        self.config = config
+        self.cache_dir = Path(cache_dir) if cache_dir else None
+        self.enable_cache = enable_cache
+        self.num_workers = max(1, num_workers)
+
+        # 模型实例
+        self._model: Optional[SentenceTransformer] = None
+        self._model_lock = threading.Lock()
+
+        # 缓存
+        self._embedding_cache: Dict[str, np.ndarray] = {}
+        self._cache_lock = threading.Lock()
+
+        # 统计
+        self._total_encoded = 0
+        self._cache_hits = 0
+        self._cache_misses = 0
+
+        logger.info(
+            f"EmbeddingManager 初始化: model={config.model_name}, "
+            f"dim={config.dimension}, workers={num_workers}"
+        )
+
+    def load_model(self) -> None:
+        """加载模型（懒加载）"""
+        if self._model is not None:
+            return
+
+        with self._model_lock:
+            # 双重检查
+            if self._model is not None:
+                return
+
+            logger.info(f"正在加载模型: {self.config.model_name}")
+
+            try:
+                # 构建模型参数
+                model_kwargs = {}
+                if self.config.cache_dir:
+                    model_kwargs["cache_folder"] = self.config.cache_dir
+
+                # 加载模型
+                self._model = SentenceTransformer(
+                    self.config.model_path,
+                    **model_kwargs,
+                )
+
+                logger.info(f"模型加载成功: {self.config.model_name}")
+
+            except Exception as e:
+                logger.error(f"模型加载失败: {e}")
+                raise
+
+    def encode(
+        self,
+        texts: Union[str, List[str]],
+        batch_size: Optional[int] = None,
+        show_progress: bool = False,
+        normalize: bool = True,
+    ) -> np.ndarray:
+        """
+        生成文本嵌入
+
+        Args:
+            texts: 文本或文本列表
+            batch_size: 批次大小（默认使用配置值）
+            show_progress: 是否显示进度条
+            normalize: 是否归一化
+
+        Returns:
+            嵌入向量 (N x D)
+        """
+        # 确保模型已加载
+        self.load_model()
+
+        # 标准化输入
+        if isinstance(texts, str):
+            texts = [texts]
+            single_input = True
+        else:
+            single_input = False
+
+        if not texts:
+            return np.zeros((0, self.config.dimension), dtype=np.float32)
+
+        # 使用配置的批次大小
+        if batch_size is None:
+            batch_size = self.config.batch_size
+
+        # 生成嵌入
+        try:
+            embeddings = self._model.encode(
+                texts,
+                batch_size=batch_size,
+                show_progress_bar=show_progress,
+                normalize_embeddings=normalize and self.config.normalization,
+                convert_to_numpy=True,
+            )
+
+            # 确保是2D数组
+            if embeddings.ndim == 1:
+                embeddings = embeddings.reshape(1, -1)
+
+            self._total_encoded += len(texts)
+
+            # 如果是单个输入，返回1D数组
+            if single_input:
+                return embeddings[0]
+
+            return embeddings
+
+        except Exception as e:
+            logger.error(f"生成嵌入失败: {e}")
+            raise
+
+    def encode_batch(
+        self,
+        texts: List[str],
+        batch_size: Optional[int] = None,
+        num_workers: Optional[int] = None,
+        show_progress: bool = False,
+    ) -> np.ndarray:
+        """
+        批量生成嵌入（多线程优化）
+
+        Args:
+            texts: 文本列表
+            batch_size: 批次大小
+            num_workers: 工作线程数（默认使用初始化时的值）
+            show_progress: 是否显示进度条
+
+        Returns:
+            嵌入向量 (N x D)
+        """
+        if not texts:
+            return np.zeros((0, self.config.dimension), dtype=np.float32)
+
+        # 单线程模式
+        num_workers = num_workers if num_workers is not None else self.num_workers
+        if num_workers == 1:
+            return self.encode(texts, batch_size=batch_size, show_progress=show_progress)
+
+        # 多线程模式
+        logger.info(f"使用 {num_workers} 个线程生成 {len(texts)} 个嵌入")
+
+        # 分批
+        batch_size = batch_size or self.config.batch_size
+        batches = [
+            texts[i:i + batch_size]
+            for i in range(0, len(texts), batch_size)
+        ]
+
+        # 多线程生成
+        all_embeddings = []
+        with ThreadPoolExecutor(max_workers=num_workers) as executor:
+            # 提交任务
+            future_to_batch = {
+                executor.submit(
+                    self.encode,
+                    batch,
+                    batch_size,
+                    False,  # 不显示进度条（多线程时会混乱）
+                ): i
+                for i, batch in enumerate(batches)
+            }
+
+            # 收集结果
+            for future in as_completed(future_to_batch):
+                batch_idx = future_to_batch[future]
+                try:
+                    embeddings = future.result()
+                    all_embeddings.append((batch_idx, embeddings))
+                except Exception as e:
+                    logger.error(f"批次 {batch_idx} 生成嵌入失败: {e}")
+                    raise
+
+        # 按顺序合并
+        all_embeddings.sort(key=lambda x: x[0])
+        final_embeddings = np.concatenate([emb for _, emb in all_embeddings], axis=0)
+
+        return final_embeddings
+
+    def encode_with_cache(
+        self,
+        texts: List[str],
+        batch_size: Optional[int] = None,
+        show_progress: bool = False,
+    ) -> np.ndarray:
+        """
+        生成嵌入（带缓存）
+
+        Args:
+            texts: 文本列表
+            batch_size: 批次大小
+            show_progress: 是否显示进度条
+
+        Returns:
+            嵌入向量 (N x D)
+        """
+        if not self.enable_cache:
+            return self.encode(texts, batch_size, show_progress)
+
+        # 分离缓存命中和未命中的文本
+        cached_embeddings = []
+        uncached_texts = []
+        uncached_indices = []
+
+        for i, text in enumerate(texts):
+            cache_key = self._get_cache_key(text)
+
+            with self._cache_lock:
+                if cache_key in self._embedding_cache:
+                    cached_embeddings.append((i, self._embedding_cache[cache_key]))
+                    self._cache_hits += 1
+                else:
+                    uncached_texts.append(text)
+                    uncached_indices.append(i)
+                    self._cache_misses += 1
+
+        # 生成未缓存的嵌入
+        if uncached_texts:
+            new_embeddings = self.encode(
+                uncached_texts,
+                batch_size,
+                show_progress,
+            )
+
+            # 更新缓存
+            with self._cache_lock:
+                for text, embedding in zip(uncached_texts, new_embeddings):
+                    cache_key = self._get_cache_key(text)
+                    self._embedding_cache[cache_key] = embedding.copy()
+
+            # 合并结果
+            for idx, embedding in zip(uncached_indices, new_embeddings):
+                cached_embeddings.append((idx, embedding))
+
+        # 按原始顺序排序
+        cached_embeddings.sort(key=lambda x: x[0])
+        final_embeddings = np.array([emb for _, emb in cached_embeddings])
+
+        return final_embeddings
+
+    def save_cache(self, cache_path: Optional[Union[str, Path]] = None) -> None:
+        """
+        保存缓存到磁盘
+
+        Args:
+            cache_path: 缓存文件路径（默认使用cache_dir/embeddings_cache.pkl）
+        """
+        if cache_path is None:
+            if self.cache_dir is None:
+                raise ValueError("未指定缓存目录")
+            cache_path = self.cache_dir / "embeddings_cache.pkl"
+
+        cache_path = Path(cache_path)
+        cache_path.parent.mkdir(parents=True, exist_ok=True)
+
+        with self._cache_lock:
+            with open(cache_path, "wb") as f:
+                pickle.dump(self._embedding_cache, f)
+
+            logger.info(f"缓存已保存: {cache_path} ({len(self._embedding_cache)} 条)")
+
+    def load_cache(self, cache_path: Optional[Union[str, Path]] = None) -> None:
+        """
+        从磁盘加载缓存
+
+        Args:
+            cache_path: 缓存文件路径（默认使用cache_dir/embeddings_cache.pkl）
+        """
+        if cache_path is None:
+            if self.cache_dir is None:
+                raise ValueError("未指定缓存目录")
+            cache_path = self.cache_dir / "embeddings_cache.pkl"
+
+        cache_path = Path(cache_path)
+        if not cache_path.exists():
+            logger.warning(f"缓存文件不存在: {cache_path}")
+            return
+
+        with self._cache_lock:
+            with open(cache_path, "rb") as f:
+                self._embedding_cache = pickle.load(f)
+
+            logger.info(f"缓存已加载: {cache_path} ({len(self._embedding_cache)} 条)")
+
+    def clear_cache(self) -> None:
+        """清空缓存"""
+        with self._cache_lock:
+            count = len(self._embedding_cache)
+            self._embedding_cache.clear()
+            logger.info(f"已清空缓存: {count} 条")
+
+    def check_model_consistency(
+        self,
+        stored_embeddings: np.ndarray,
+        sample_texts: List[str] = None,
+    ) -> Tuple[bool, str]:
+        """
+        检查模型一致性
+
+        Args:
+            stored_embeddings: 存储的嵌入向量
+            sample_texts: 样本文本（用于重新生成对比）
+
+        Returns:
+            (是否一致, 详细信息)
+        """
+        # 检查维度
+        if stored_embeddings.shape[1] != self.config.dimension:
+            return False, f"维度不匹配: 期望 {self.config.dimension}, 实际 {stored_embeddings.shape[1]}"
+
+        # 如果提供了样本文本，重新生成并比较
+        if sample_texts:
+            try:
+                new_embeddings = self.encode(sample_texts[:5])  # 只比较前5个
+
+                # 计算相似度
+                similarities = np.dot(
+                    stored_embeddings[:5],
+                    new_embeddings.T,
+                ).diagonal()
+
+                # 检查相似度
+                if np.mean(similarities) < 0.95:
+                    return False, f"模型可能已更改，平均相似度: {np.mean(similarities):.3f}"
+
+                return True, f"模型一致，平均相似度: {np.mean(similarities):.3f}"
+
+            except Exception as e:
+                return False, f"一致性检查失败: {e}"
+
+        return True, "维度匹配"
+
+    def get_model_info(self) -> Dict[str, Any]:
+        """
+        获取模型信息
+
+        Returns:
+            模型信息字典
+        """
+        return {
+            "model_name": self.config.model_name,
+            "dimension": self.config.dimension,
+            "max_seq_length": self.config.max_seq_length,
+            "batch_size": self.config.batch_size,
+            "normalization": self.config.normalization,
+            "pooling": self.config.pooling,
+            "model_loaded": self._model is not None,
+            "cache_enabled": self.enable_cache,
+            "cache_size": len(self._embedding_cache),
+            "total_encoded": self._total_encoded,
+            "cache_hits": self._cache_hits,
+            "cache_misses": self._cache_misses,
+        }
+
+    def get_embedding_dimension(self) -> int:
+        """获取嵌入维度"""
+        return self.config.dimension
+
+    def _get_cache_key(self, text: str) -> str:
+        """
+        生成缓存键
+
+        Args:
+            text: 文本内容
+
+        Returns:
+            缓存键（SHA256哈希）
+        """
+        return hashlib.sha256(text.encode("utf-8")).hexdigest()
+
+    @property
+    def is_model_loaded(self) -> bool:
+        """模型是否已加载"""
+        return self._model is not None
+
+    @property
+    def cache_hit_rate(self) -> float:
+        """缓存命中率"""
+        total = self._cache_hits + self._cache_misses
+        if total == 0:
+            return 0.0
+        return self._cache_hits / total
+
+    def __repr__(self) -> str:
+        return (
+            f"EmbeddingManager(model={self.config.model_name}, "
+            f"dim={self.config.dimension}, "
+            f"loaded={self.is_model_loaded}, "
+            f"cache={len(self._embedding_cache)})"
+        )
+
+
+
+
+def create_embedding_manager_from_config(
+    model_name: str,
+    model_path: str,
+    dimension: int,
+    cache_dir: Optional[Union[str, Path]] = None,
+    enable_cache: bool = True,
+    num_workers: int = 1,
+    **config_kwargs,
+) -> EmbeddingManager:
+    """
+    从自定义配置创建嵌入管理器
+
+    Args:
+        model_name: 模型名称
+        model_path: HuggingFace模型路径
+        dimension: 输出维度
+        cache_dir: 缓存目录
+        enable_cache: 是否启用缓存
+        num_workers: 工作线程数
+        **config_kwargs: 其他配置参数
+
+    Returns:
+        嵌入管理器实例
+    """
+    # 创建自定义配置
+    config = get_custom_config(
+        model_name=model_name,
+        model_path=model_path,
+        dimension=dimension,
+        cache_dir=cache_dir,
+        **config_kwargs,
+    )
+
+    # 创建管理器
+    return EmbeddingManager(
+        config=config,
+        cache_dir=cache_dir,
+        enable_cache=enable_cache,
+        num_workers=num_workers,
+    )

plugins/A_memorix/core/embedding/presets.py

@@ -0,0 +1,72 @@
+"""
+嵌入模型配置模块
+"""
+
+from dataclasses import dataclass
+from typing import Optional, Dict, Any, Union
+from pathlib import Path
+
+
+@dataclass
+class EmbeddingModelConfig:
+    """
+    嵌入模型配置
+
+    属性：
+        model_name: 模型描述名称
+        model_path: 实际加载路径（Local or HF）
+        dimension: 嵌入向量维度
+        max_seq_length: 最大序列长度
+        batch_size: 编码批次大小
+        model_size_mb: 估计显存占用
+        description: 模型说明
+        normalization: 是否自动归一化
+        pooling: 池化策略 (mean, cls, max)
+        cache_dir: 模型缓存目录
+    """
+
+    model_name: str
+    model_path: str
+    dimension: int
+    max_seq_length: int = 512
+    batch_size: int = 32
+    model_size_mb: int = 100
+    description: str = ""
+    normalization: bool = True
+    pooling: str = "mean"
+    cache_dir: Optional[Union[str, Path]] = None
+
+
+def validate_config_compatibility(
+    config1: EmbeddingModelConfig, config2: EmbeddingModelConfig
+) -> bool:
+    """检查两个配置是否兼容（主要看维度）"""
+    return config1.dimension == config2.dimension
+
+
+def are_models_compatible(
+    config1: EmbeddingModelConfig, config2: EmbeddingModelConfig
+) -> bool:
+    """检查模型是否完全相同（用于热切换判断）"""
+    return (
+        config1.model_path == config2.model_path
+        and config1.dimension == config2.dimension
+        and config1.pooling == config2.pooling
+    )
+
+
+def get_custom_config(
+    model_name: str,
+    model_path: str,
+    dimension: int,
+    cache_dir: Optional[Union[str, Path]] = None,
+    **kwargs,
+) -> EmbeddingModelConfig:
+    """创建自定义模型配置"""
+    return EmbeddingModelConfig(
+        model_name=model_name,
+        model_path=model_path,
+        dimension=dimension,
+        cache_dir=cache_dir,
+        **kwargs,
+    )

plugins/A_memorix/core/retrieval/__init__.py

@@ -0,0 +1,54 @@
+"""检索模块 - 双路检索与排序"""
+
+from .dual_path import (
+    DualPathRetriever,
+    RetrievalStrategy,
+    RetrievalResult,
+    DualPathRetrieverConfig,
+    TemporalQueryOptions,
+    FusionConfig,
+    RelationIntentConfig,
+)
+from .pagerank import (
+    PersonalizedPageRank,
+    PageRankConfig,
+    create_ppr_from_graph,
+)
+from .threshold import (
+    DynamicThresholdFilter,
+    ThresholdMethod,
+    ThresholdConfig,
+)
+from .sparse_bm25 import (
+    SparseBM25Index,
+    SparseBM25Config,
+)
+from .graph_relation_recall import (
+    GraphRelationRecallConfig,
+    GraphRelationRecallService,
+)
+
+__all__ = [
+    # DualPathRetriever
+    "DualPathRetriever",
+    "RetrievalStrategy",
+    "RetrievalResult",
+    "DualPathRetrieverConfig",
+    "TemporalQueryOptions",
+    "FusionConfig",
+    "RelationIntentConfig",
+    # PersonalizedPageRank
+    "PersonalizedPageRank",
+    "PageRankConfig",
+    "create_ppr_from_graph",
+    # DynamicThresholdFilter
+    "DynamicThresholdFilter",
+    "ThresholdMethod",
+    "ThresholdConfig",
+    # Sparse BM25
+    "SparseBM25Index",
+    "SparseBM25Config",
+    # Graph relation recall
+    "GraphRelationRecallConfig",
+    "GraphRelationRecallService",
+]

plugins/A_memorix/core/retrieval/graph_relation_recall.py

@@ -0,0 +1,272 @@
+"""Graph-assisted relation candidate recall for relation-oriented queries."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Any, Dict, List, Optional, Sequence, Set
+
+from src.common.logger import get_logger
+
+logger = get_logger("A_Memorix.GraphRelationRecall")
+
+
+@dataclass
+class GraphRelationRecallConfig:
+    """Configuration for controlled graph relation recall."""
+
+    enabled: bool = True
+    candidate_k: int = 24
+    max_hop: int = 1
+    allow_two_hop_pair: bool = True
+    max_paths: int = 4
+
+    def __post_init__(self) -> None:
+        self.enabled = bool(self.enabled)
+        self.candidate_k = max(1, int(self.candidate_k))
+        self.max_hop = max(1, int(self.max_hop))
+        self.allow_two_hop_pair = bool(self.allow_two_hop_pair)
+        self.max_paths = max(1, int(self.max_paths))
+
+
+@dataclass
+class GraphRelationCandidate:
+    """A graph-derived relation candidate before retriever-side fusion."""
+
+    hash_value: str
+    subject: str
+    predicate: str
+    object: str
+    confidence: float
+    graph_seed_entities: List[str]
+    graph_hops: int
+    graph_candidate_type: str
+    supporting_paragraph_count: int
+
+    def to_payload(self) -> Dict[str, Any]:
+        content = f"{self.subject} {self.predicate} {self.object}"
+        return {
+            "hash": self.hash_value,
+            "content": content,
+            "subject": self.subject,
+            "predicate": self.predicate,
+            "object": self.object,
+            "confidence": self.confidence,
+            "graph_seed_entities": list(self.graph_seed_entities),
+            "graph_hops": int(self.graph_hops),
+            "graph_candidate_type": self.graph_candidate_type,
+            "supporting_paragraph_count": int(self.supporting_paragraph_count),
+        }
+
+
+class GraphRelationRecallService:
+    """Collect relation candidates from the entity graph in a controlled way."""
+
+    def __init__(
+        self,
+        *,
+        graph_store: Any,
+        metadata_store: Any,
+        config: Optional[GraphRelationRecallConfig] = None,
+    ) -> None:
+        self.graph_store = graph_store
+        self.metadata_store = metadata_store
+        self.config = config or GraphRelationRecallConfig()
+
+    def recall(
+        self,
+        *,
+        seed_entities: Sequence[str],
+    ) -> List[GraphRelationCandidate]:
+        if not self.config.enabled:
+            return []
+        if self.graph_store is None or self.metadata_store is None:
+            return []
+
+        seeds = self._normalize_seed_entities(seed_entities)
+        if not seeds:
+            return []
+
+        seen_hashes: Set[str] = set()
+        candidates: List[GraphRelationCandidate] = []
+
+        if len(seeds) >= 2:
+            self._collect_direct_pair_candidates(
+                seed_a=seeds[0],
+                seed_b=seeds[1],
+                seen_hashes=seen_hashes,
+                out=candidates,
+            )
+            if (
+                len(candidates) < 3
+                and self.config.allow_two_hop_pair
+                and len(candidates) < self.config.candidate_k
+            ):
+                self._collect_two_hop_pair_candidates(
+                    seed_a=seeds[0],
+                    seed_b=seeds[1],
+                    seen_hashes=seen_hashes,
+                    out=candidates,
+                )
+        else:
+            self._collect_one_hop_seed_candidates(
+                seed=seeds[0],
+                seen_hashes=seen_hashes,
+                out=candidates,
+            )
+
+        return candidates[: self.config.candidate_k]
+
+    def _normalize_seed_entities(self, seed_entities: Sequence[str]) -> List[str]:
+        out: List[str] = []
+        seen = set()
+        for raw in list(seed_entities)[:2]:
+            resolved = None
+            try:
+                resolved = self.graph_store.find_node(str(raw), ignore_case=True)
+            except Exception:
+                resolved = None
+            if not resolved:
+                continue
+            canon = str(resolved).strip().lower()
+            if not canon or canon in seen:
+                continue
+            seen.add(canon)
+            out.append(str(resolved))
+        return out
+
+    def _collect_direct_pair_candidates(
+        self,
+        *,
+        seed_a: str,
+        seed_b: str,
+        seen_hashes: Set[str],
+        out: List[GraphRelationCandidate],
+    ) -> None:
+        relation_hashes = []
+        relation_hashes.extend(self.graph_store.get_relation_hashes_for_edge(seed_a, seed_b))
+        relation_hashes.extend(self.graph_store.get_relation_hashes_for_edge(seed_b, seed_a))
+        self._append_relation_hashes(
+            relation_hashes=relation_hashes,
+            seen_hashes=seen_hashes,
+            out=out,
+            candidate_type="direct_pair",
+            graph_hops=1,
+            graph_seed_entities=[seed_a, seed_b],
+        )
+
+    def _collect_two_hop_pair_candidates(
+        self,
+        *,
+        seed_a: str,
+        seed_b: str,
+        seen_hashes: Set[str],
+        out: List[GraphRelationCandidate],
+    ) -> None:
+        try:
+            paths = self.graph_store.find_paths(
+                seed_a,
+                seed_b,
+                max_depth=2,
+                max_paths=self.config.max_paths,
+            )
+        except Exception as e:
+            logger.debug("graph two-hop recall skipped: %s", e)
+            return
+
+        for path_nodes in paths:
+            if len(out) >= self.config.candidate_k:
+                break
+            if not isinstance(path_nodes, Sequence) or len(path_nodes) < 3:
+                continue
+            if len(path_nodes) != 3:
+                continue
+            for idx in range(len(path_nodes) - 1):
+                if len(out) >= self.config.candidate_k:
+                    break
+                u = str(path_nodes[idx])
+                v = str(path_nodes[idx + 1])
+                relation_hashes = []
+                relation_hashes.extend(self.graph_store.get_relation_hashes_for_edge(u, v))
+                relation_hashes.extend(self.graph_store.get_relation_hashes_for_edge(v, u))
+                self._append_relation_hashes(
+                    relation_hashes=relation_hashes,
+                    seen_hashes=seen_hashes,
+                    out=out,
+                    candidate_type="two_hop_pair",
+                    graph_hops=2,
+                    graph_seed_entities=[seed_a, seed_b],
+                )
+
+    def _collect_one_hop_seed_candidates(
+        self,
+        *,
+        seed: str,
+        seen_hashes: Set[str],
+        out: List[GraphRelationCandidate],
+    ) -> None:
+        try:
+            relation_hashes = self.graph_store.get_incident_relation_hashes(
+                seed,
+                limit=self.config.candidate_k,
+            )
+        except Exception as e:
+            logger.debug("graph one-hop recall skipped: %s", e)
+            return
+        self._append_relation_hashes(
+            relation_hashes=relation_hashes,
+            seen_hashes=seen_hashes,
+            out=out,
+            candidate_type="one_hop_seed",
+            graph_hops=min(1, self.config.max_hop),
+            graph_seed_entities=[seed],
+        )
+
+    def _append_relation_hashes(
+        self,
+        *,
+        relation_hashes: Sequence[str],
+        seen_hashes: Set[str],
+        out: List[GraphRelationCandidate],
+        candidate_type: str,
+        graph_hops: int,
+        graph_seed_entities: Sequence[str],
+    ) -> None:
+        for relation_hash in sorted({str(h) for h in relation_hashes if str(h).strip()}):
+            if len(out) >= self.config.candidate_k:
+                break
+            if relation_hash in seen_hashes:
+                continue
+            candidate = self._build_candidate(
+                relation_hash=relation_hash,
+                candidate_type=candidate_type,
+                graph_hops=graph_hops,
+                graph_seed_entities=graph_seed_entities,
+            )
+            if candidate is None:
+                continue
+            seen_hashes.add(relation_hash)
+            out.append(candidate)
+
+    def _build_candidate(
+        self,
+        *,
+        relation_hash: str,
+        candidate_type: str,
+        graph_hops: int,
+        graph_seed_entities: Sequence[str],
+    ) -> Optional[GraphRelationCandidate]:
+        relation = self.metadata_store.get_relation(relation_hash)
+        if relation is None:
+            return None
+        supporting_paragraphs = self.metadata_store.get_paragraphs_by_relation(relation_hash)
+        return GraphRelationCandidate(
+            hash_value=relation_hash,
+            subject=str(relation.get("subject", "")),
+            predicate=str(relation.get("predicate", "")),
+            object=str(relation.get("object", "")),
+            confidence=float(relation.get("confidence", 1.0) or 1.0),
+            graph_seed_entities=[str(x) for x in graph_seed_entities],
+            graph_hops=int(graph_hops),
+            graph_candidate_type=str(candidate_type),
+            supporting_paragraph_count=len(supporting_paragraphs),
+        )

plugins/A_memorix/core/retrieval/pagerank.py

@@ -0,0 +1,482 @@
+"""
+Personalized PageRank实现
+
+提供个性化的图节点排序功能。
+"""
+
+from typing import Dict, List, Optional, Tuple, Union, Any
+from dataclasses import dataclass
+import numpy as np
+
+from src.common.logger import get_logger
+from ..storage import GraphStore
+from ..utils.matcher import AhoCorasick
+
+logger = get_logger("A_Memorix.PersonalizedPageRank")
+
+
+@dataclass
+class PageRankConfig:
+    """
+    PageRank配置
+
+    属性：
+        alpha: 阻尼系数（0-1之间）
+        max_iter: 最大迭代次数
+        tol: 收敛阈值
+        normalize: 是否归一化结果
+        min_iterations: 最小迭代次数
+    """
+
+    alpha: float = 0.85
+    max_iter: int = 100
+    tol: float = 1e-6
+    normalize: bool = True
+    min_iterations: int = 20
+
+    def __post_init__(self):
+        """验证配置"""
+        if not 0 <= self.alpha < 1:
+            raise ValueError(f"alpha必须在[0, 1)之间: {self.alpha}")
+
+        if self.max_iter <= 0:
+            raise ValueError(f"max_iter必须大于0: {self.max_iter}")
+
+        if self.tol <= 0:
+            raise ValueError(f"tol必须大于0: {self.tol}")
+
+        if self.min_iterations < 0:
+            raise ValueError(f"min_iterations必须大于等于0: {self.min_iterations}")
+
+        if self.min_iterations >= self.max_iter:
+            raise ValueError(f"min_iterations必须小于max_iter")
+
+
+class PersonalizedPageRank:
+    """
+    Personalized PageRank计算器
+
+    功能：
+    - 个性化向量支持
+    - 快速收敛检测
+    - 结果归一化
+    - 批量计算
+    - 统计信息
+
+    参数：
+        graph_store: 图存储
+        config: PageRank配置
+    """
+
+    def __init__(
+        self,
+        graph_store: GraphStore,
+        config: Optional[PageRankConfig] = None,
+    ):
+        """
+        初始化PPR计算器
+
+        Args:
+            graph_store: 图存储
+            config: PageRank配置
+        """
+        self.graph_store = graph_store
+        self.config = config or PageRankConfig()
+
+        # 统计信息
+        self._total_computations = 0
+        self._total_iterations = 0
+        self._convergence_history: List[int] = []
+
+        logger.info(
+            f"PersonalizedPageRank 初始化: "
+            f"alpha={self.config.alpha}, "
+            f"max_iter={self.config.max_iter}"
+        )
+
+        # 缓存 Aho-Corasick 匹配器
+        self._ac_matcher: Optional[AhoCorasick] = None
+        self._ac_nodes_count = 0
+
+    def compute(
+        self,
+        personalization: Optional[Dict[str, float]] = None,
+        alpha: Optional[float] = None,
+        max_iter: Optional[int] = None,
+        normalize: Optional[bool] = None,
+    ) -> Dict[str, float]:
+        """
+        计算Personalized PageRank
+
+        Args:
+            personalization: 个性化向量 {节点名: 权重}
+            alpha: 阻尼系数（覆盖配置值）
+            max_iter: 最大迭代次数（覆盖配置值）
+            normalize: 是否归一化（覆盖配置值）
+
+        Returns:
+            节点PageRank值字典 {节点名: 分数}
+        """
+        # 使用覆盖值或配置值
+        alpha = alpha if alpha is not None else self.config.alpha
+        max_iter = max_iter if max_iter is not None else self.config.max_iter
+        normalize = normalize if normalize is not None else self.config.normalize
+
+        # 调用GraphStore的compute_pagerank
+        scores = self.graph_store.compute_pagerank(
+            personalization=personalization,
+            alpha=alpha,
+            max_iter=max_iter,
+            tol=self.config.tol,
+        )
+
+        # 归一化（如果需要）
+        if normalize and scores:
+            total = sum(scores.values())
+            if total > 0:
+                scores = {node: score / total for node, score in scores.items()}
+
+        # 更新统计
+        self._total_computations += 1
+
+        logger.debug(
+            f"PPR计算完成: {len(scores)} 个节点, "
+            f"personalization_nodes={len(personalization) if personalization else 0}"
+        )
+
+        return scores
+
+    def compute_batch(
+        self,
+        personalization_list: List[Dict[str, float]],
+        normalize: bool = True,
+    ) -> List[Dict[str, float]]:
+        """
+        批量计算PPR
+
+        Args:
+            personalization_list: 个性化向量列表
+            normalize: 是否归一化
+
+        Returns:
+            PageRank值字典列表
+        """
+        results = []
+
+        for i, personalization in enumerate(personalization_list):
+            logger.debug(f"计算第 {i+1}/{len(personalization_list)} 个PPR")
+            scores = self.compute(personalization=personalization, normalize=normalize)
+            results.append(scores)
+
+        return results
+
+    def compute_for_entities(
+        self,
+        entities: List[str],
+        weights: Optional[List[float]] = None,
+        normalize: bool = True,
+    ) -> Dict[str, float]:
+        """
+        为实体列表计算PPR
+
+        Args:
+            entities: 实体列表
+            weights: 权重列表（默认均匀权重）
+            normalize: 是否归一化
+
+        Returns:
+            PageRank值字典
+        """
+        if not entities:
+            logger.warning("实体列表为空，返回均匀PPR")
+            return self.compute(personalization=None, normalize=normalize)
+
+        # 构建个性化向量
+        if weights is None:
+            weights = [1.0] * len(entities)
+
+        if len(weights) != len(entities):
+            raise ValueError(f"权重数量与实体数量不匹配: {len(weights)} vs {len(entities)}")
+
+        personalization = {entity: weight for entity, weight in zip(entities, weights)}
+
+        return self.compute(personalization=personalization, normalize=normalize)
+
+    def compute_for_query(
+        self,
+        query: str,
+        entity_extractor: Optional[callable] = None,
+        normalize: bool = True,
+    ) -> Dict[str, float]:
+        """
+        为查询计算PPR
+
+        Args:
+            query: 查询文本
+            entity_extractor: 实体提取函数（可选）
+            normalize: 是否归一化
+
+        Returns:
+            PageRank值字典
+        """
+        # 提取实体
+        if entity_extractor is not None:
+            entities = entity_extractor(query)
+        else:
+            # 简单实现：基于图中的节点匹配
+            entities = self._extract_entities_from_query(query)
+
+        if not entities:
+            logger.debug(f"未从查询中提取到实体: '{query}'")
+            return self.compute(personalization=None, normalize=normalize)
+
+        # 计算PPR
+        return self.compute_for_entities(entities, normalize=normalize)
+
+    def rank_nodes(
+        self,
+        scores: Dict[str, float],
+        top_k: Optional[int] = None,
+        min_score: float = 0.0,
+    ) -> List[Tuple[str, float]]:
+        """
+        对节点排序
+
+        Args:
+            scores: PageRank分数字典
+            top_k: 返回前k个节点（None表示全部）
+            min_score: 最小分数阈值
+
+        Returns:
+            排序后的节点列表 [(节点名, 分数), ...]
+        """
+        # 过滤低分节点
+        filtered = [(node, score) for node, score in scores.items() if score >= min_score]
+
+        # 按分数降序排序
+        sorted_nodes = sorted(filtered, key=lambda x: x[1], reverse=True)
+
+        # 返回top_k
+        if top_k is not None:
+            sorted_nodes = sorted_nodes[:top_k]
+
+        return sorted_nodes
+
+    def get_personalization_vector(
+        self,
+        nodes: List[str],
+        method: str = "uniform",
+    ) -> Dict[str, float]:
+        """
+        生成个性化向量
+
+        Args:
+            nodes: 节点列表
+            method: 生成方法
+                - "uniform": 均匀权重
+                - "degree": 按度数加权
+                - "inverse_degree": 按度数反比加权
+
+        Returns:
+            个性化向量 {节点名: 权重}
+        """
+        if not nodes:
+            return {}
+
+        if method == "uniform":
+            # 均匀权重
+            weight = 1.0 / len(nodes)
+            return {node: weight for node in nodes}
+
+        elif method == "degree":
+            # 按度数加权
+            node_degrees = {}
+            for node in nodes:
+                neighbors = self.graph_store.get_neighbors(node)
+                node_degrees[node] = len(neighbors)
+
+            total_degree = sum(node_degrees.values())
+            if total_degree > 0:
+                return {node: degree / total_degree for node, degree in node_degrees.items()}
+            else:
+                return {node: 1.0 / len(nodes) for node in nodes}
+
+        elif method == "inverse_degree":
+            # 按度数反比加权
+            node_degrees = {}
+            for node in nodes:
+                neighbors = self.graph_store.get_neighbors(node)
+                node_degrees[node] = len(neighbors)
+
+            # 反度数
+            inv_degrees = {node: 1.0 / (degree + 1) for node, degree in node_degrees.items()}
+            total_inv = sum(inv_degrees.values())
+
+            if total_inv > 0:
+                return {node: inv / total_inv for node, inv in inv_degrees.items()}
+            else:
+                return {node: 1.0 / len(nodes) for node in nodes}
+
+        else:
+            raise ValueError(f"不支持的个性化向量生成方法: {method}")
+
+    def compare_scores(
+        self,
+        scores1: Dict[str, float],
+        scores2: Dict[str, float],
+    ) -> Dict[str, Dict[str, float]]:
+        """
+        比较两组PPR分数
+
+        Args:
+            scores1: 第一组分数
+            scores2: 第二组分数
+
+        Returns:
+            比较结果 {
+                "common_nodes": {节点: (score1, score2)},
+                "only_in_1": {节点: score1},
+                "only_in_2": {节点: score2},
+            }
+        """
+        common_nodes = {}
+        only_in_1 = {}
+        only_in_2 = {}
+
+        all_nodes = set(scores1.keys()) | set(scores2.keys())
+
+        for node in all_nodes:
+            if node in scores1 and node in scores2:
+                common_nodes[node] = (scores1[node], scores2[node])
+            elif node in scores1:
+                only_in_1[node] = scores1[node]
+            else:
+                only_in_2[node] = scores2[node]
+
+        return {
+            "common_nodes": common_nodes,
+            "only_in_1": only_in_1,
+            "only_in_2": only_in_2,
+        }
+
+    def get_statistics(self) -> Dict[str, Any]:
+        """
+        获取统计信息
+
+        Returns:
+            统计信息字典
+        """
+        avg_iterations = (
+            self._total_iterations / self._total_computations
+            if self._total_computations > 0
+            else 0
+        )
+
+        return {
+            "config": {
+                "alpha": self.config.alpha,
+                "max_iter": self.config.max_iter,
+                "tol": self.config.tol,
+                "normalize": self.config.normalize,
+                "min_iterations": self.config.min_iterations,
+            },
+            "statistics": {
+                "total_computations": self._total_computations,
+                "total_iterations": self._total_iterations,
+                "avg_iterations": avg_iterations,
+                "convergence_history": self._convergence_history.copy(),
+            },
+            "graph": {
+                "num_nodes": self.graph_store.num_nodes,
+                "num_edges": self.graph_store.num_edges,
+            },
+        }
+
+    def reset_statistics(self) -> None:
+        """重置统计信息"""
+        self._total_computations = 0
+        self._total_iterations = 0
+        self._convergence_history.clear()
+        logger.info("统计信息已重置")
+
+    def _extract_entities_from_query(self, query: str) -> List[str]:
+        """
+        从查询中提取实体（简化实现）
+
+        Args:
+            query: 查询文本
+
+        Returns:
+            实体列表
+        """
+        # 获取所有节点
+        all_nodes = self.graph_store.get_nodes()
+        if not all_nodes:
+            return []
+
+        # 检查是否需要更新 Aho-Corasick 匹配器
+        if self._ac_matcher is None or self._ac_nodes_count != len(all_nodes):
+            self._ac_matcher = AhoCorasick()
+            for node in all_nodes:
+                # 统一转为小写进行不区分大小写匹配
+                self._ac_matcher.add_pattern(node.lower())
+            self._ac_matcher.build()
+            self._ac_nodes_count = len(all_nodes)
+
+        # 执行匹配
+        query_lower = query.lower()
+        stats = self._ac_matcher.find_all(query_lower)
+        
+        # 转换回原始的大小写（这里简化为从 all_nodes 中找，或者 AC 存原始值）
+        # 为了简单，AC 中 add_pattern 存的是小写
+        # 我们需要一个映射：小写 -> 原始
+        node_map = {node.lower(): node for node in all_nodes}
+        entities = [node_map[low_name] for low_name in stats.keys()]
+
+        return entities
+
+    @property
+    def num_computations(self) -> int:
+        """计算次数"""
+        return self._total_computations
+
+    @property
+    def avg_iterations(self) -> float:
+        """平均迭代次数"""
+        if self._total_computations == 0:
+            return 0.0
+        return self._total_iterations / self._total_computations
+
+    def __repr__(self) -> str:
+        return (
+            f"PersonalizedPageRank("
+            f"alpha={self.config.alpha}, "
+            f"computations={self._total_computations})"
+        )
+
+
+def create_ppr_from_graph(
+    graph_store: GraphStore,
+    alpha: float = 0.85,
+    max_iter: int = 100,
+) -> PersonalizedPageRank:
+    """
+    从图存储创建PPR计算器
+
+    Args:
+        graph_store: 图存储
+        alpha: 阻尼系数
+        max_iter: 最大迭代次数
+
+    Returns:
+        PPR计算器实例
+    """
+    config = PageRankConfig(
+        alpha=alpha,
+        max_iter=max_iter,
+    )
+
+    return PersonalizedPageRank(
+        graph_store=graph_store,
+        config=config,
+    )

plugins/A_memorix/core/retrieval/sparse_bm25.py

@@ -0,0 +1,402 @@
+"""
+稀疏检索组件（FTS5 + BM25）
+
+支持：
+- 懒加载索引连接
+- jieba / char n-gram 分词
+- 可卸载并收缩 SQLite 内存缓存
+"""
+
+from __future__ import annotations
+
+import re
+import sqlite3
+from dataclasses import dataclass
+from typing import Any, Dict, List, Optional
+
+from src.common.logger import get_logger
+from ..storage import MetadataStore
+
+logger = get_logger("A_Memorix.SparseBM25")
+
+try:
+    import jieba  # type: ignore
+
+    HAS_JIEBA = True
+except Exception:
+    HAS_JIEBA = False
+    jieba = None
+
+
+@dataclass
+class SparseBM25Config:
+    """BM25 稀疏检索配置。"""
+
+    enabled: bool = True
+    backend: str = "fts5"
+    lazy_load: bool = True
+    mode: str = "auto"  # auto | fallback_only | hybrid
+    tokenizer_mode: str = "jieba"  # jieba | mixed | char_2gram
+    jieba_user_dict: str = ""
+    char_ngram_n: int = 2
+    candidate_k: int = 80
+    max_doc_len: int = 2000
+    enable_ngram_fallback_index: bool = True
+    enable_like_fallback: bool = False
+    enable_relation_sparse_fallback: bool = True
+    relation_candidate_k: int = 60
+    relation_max_doc_len: int = 512
+    unload_on_disable: bool = True
+    shrink_memory_on_unload: bool = True
+
+    def __post_init__(self) -> None:
+        self.backend = str(self.backend or "fts5").strip().lower()
+        self.mode = str(self.mode or "auto").strip().lower()
+        self.tokenizer_mode = str(self.tokenizer_mode or "jieba").strip().lower()
+        self.char_ngram_n = max(1, int(self.char_ngram_n))
+        self.candidate_k = max(1, int(self.candidate_k))
+        self.max_doc_len = max(0, int(self.max_doc_len))
+        self.relation_candidate_k = max(1, int(self.relation_candidate_k))
+        self.relation_max_doc_len = max(0, int(self.relation_max_doc_len))
+        if self.backend != "fts5":
+            raise ValueError(f"sparse.backend 暂仅支持 fts5: {self.backend}")
+        if self.mode not in {"auto", "fallback_only", "hybrid"}:
+            raise ValueError(f"sparse.mode 非法: {self.mode}")
+        if self.tokenizer_mode not in {"jieba", "mixed", "char_2gram"}:
+            raise ValueError(f"sparse.tokenizer_mode 非法: {self.tokenizer_mode}")
+
+
+class SparseBM25Index:
+    """
+    基于 SQLite FTS5 的 BM25 检索适配层。
+    """
+
+    def __init__(
+        self,
+        metadata_store: MetadataStore,
+        config: Optional[SparseBM25Config] = None,
+    ):
+        self.metadata_store = metadata_store
+        self.config = config or SparseBM25Config()
+        self._conn: Optional[sqlite3.Connection] = None
+        self._loaded: bool = False
+        self._jieba_dict_loaded: bool = False
+
+    @property
+    def loaded(self) -> bool:
+        return self._loaded and self._conn is not None
+
+    def ensure_loaded(self) -> bool:
+        """按需加载 FTS 连接与索引。"""
+        if not self.config.enabled:
+            return False
+        if self.loaded:
+            return True
+
+        db_path = self.metadata_store.get_db_path()
+        conn = sqlite3.connect(
+            str(db_path),
+            check_same_thread=False,
+            timeout=30.0,
+        )
+        conn.row_factory = sqlite3.Row
+        conn.execute("PRAGMA journal_mode=WAL")
+        conn.execute("PRAGMA synchronous=NORMAL")
+        conn.execute("PRAGMA temp_store=MEMORY")
+
+        if not self.metadata_store.ensure_fts_schema(conn=conn):
+            conn.close()
+            return False
+        self.metadata_store.ensure_fts_backfilled(conn=conn)
+        # 关系稀疏检索按独立开关加载，避免不必要的初始化开销。
+        if self.config.enable_relation_sparse_fallback:
+            self.metadata_store.ensure_relations_fts_schema(conn=conn)
+            self.metadata_store.ensure_relations_fts_backfilled(conn=conn)
+        if self.config.enable_ngram_fallback_index:
+            self.metadata_store.ensure_paragraph_ngram_schema(conn=conn)
+            self.metadata_store.ensure_paragraph_ngram_backfilled(
+                n=self.config.char_ngram_n,
+                conn=conn,
+            )
+
+        self._conn = conn
+        self._loaded = True
+        self._prepare_tokenizer()
+        logger.info(
+            "SparseBM25Index loaded: backend=fts5, tokenizer=%s, mode=%s",
+            self.config.tokenizer_mode,
+            self.config.mode,
+        )
+        return True
+
+    def _prepare_tokenizer(self) -> None:
+        if self._jieba_dict_loaded:
+            return
+        if self.config.tokenizer_mode not in {"jieba", "mixed"}:
+            return
+        if not HAS_JIEBA:
+            logger.warning("jieba 不可用，tokenizer 将退化为 char n-gram")
+            return
+        user_dict = str(self.config.jieba_user_dict or "").strip()
+        if user_dict:
+            try:
+                jieba.load_userdict(user_dict)  # type: ignore[union-attr]
+                logger.info("已加载 jieba 用户词典: %s", user_dict)
+            except Exception as e:
+                logger.warning("加载 jieba 用户词典失败: %s", e)
+        self._jieba_dict_loaded = True
+
+    def _tokenize_jieba(self, text: str) -> List[str]:
+        if not HAS_JIEBA:
+            return []
+        try:
+            tokens = list(jieba.cut_for_search(text))  # type: ignore[union-attr]
+            return [t.strip().lower() for t in tokens if t and t.strip()]
+        except Exception:
+            return []
+
+    def _tokenize_char_ngram(self, text: str, n: int) -> List[str]:
+        compact = re.sub(r"\s+", "", text.lower())
+        if not compact:
+            return []
+        if len(compact) < n:
+            return [compact]
+        return [compact[i : i + n] for i in range(0, len(compact) - n + 1)]
+
+    def _tokenize(self, text: str) -> List[str]:
+        text = str(text or "").strip()
+        if not text:
+            return []
+
+        mode = self.config.tokenizer_mode
+        if mode == "jieba":
+            tokens = self._tokenize_jieba(text)
+            if tokens:
+                return list(dict.fromkeys(tokens))
+            return self._tokenize_char_ngram(text, self.config.char_ngram_n)
+
+        if mode == "mixed":
+            toks = self._tokenize_jieba(text)
+            toks.extend(self._tokenize_char_ngram(text, self.config.char_ngram_n))
+            return list(dict.fromkeys([t for t in toks if t]))
+
+        return list(dict.fromkeys(self._tokenize_char_ngram(text, self.config.char_ngram_n)))
+
+    def _build_match_query(self, tokens: List[str]) -> str:
+        safe_tokens: List[str] = []
+        for token in tokens:
+            t = token.replace('"', '""').strip()
+            if not t:
+                continue
+            safe_tokens.append(f'"{t}"')
+        if not safe_tokens:
+            return ""
+        # 采用 OR 提升召回，再交由 RRF 和阈值做稳健排序。
+        return " OR ".join(safe_tokens[:64])
+
+    def _fallback_substring_search(
+        self,
+        tokens: List[str],
+        limit: int,
+    ) -> List[Dict[str, Any]]:
+        """
+        当 FTS5 因分词不一致召回为空时，退化为子串匹配召回。
+
+        说明：
+        - FTS 索引当前采用 unicode61 tokenizer。
+        - 若查询 token 来源为 char n-gram 或中文词元，可能与索引 token 不一致。
+        - 这里使用 SQL LIKE 做兜底，按命中 token 覆盖度打分。
+        """
+        if not tokens:
+            return []
+
+        # 去重并裁剪 token 数量，避免生成超长 SQL。
+        uniq_tokens = [t for t in dict.fromkeys(tokens) if t]
+        uniq_tokens = uniq_tokens[:32]
+        if not uniq_tokens:
+            return []
+
+        if self.config.enable_ngram_fallback_index:
+            try:
+                # 允许运行时切换开关后按需补齐 schema/回填。
+                self.metadata_store.ensure_paragraph_ngram_schema(conn=self._conn)
+                self.metadata_store.ensure_paragraph_ngram_backfilled(
+                    n=self.config.char_ngram_n,
+                    conn=self._conn,
+                )
+                rows = self.metadata_store.ngram_search_paragraphs(
+                    tokens=uniq_tokens,
+                    limit=limit,
+                    max_doc_len=self.config.max_doc_len,
+                    conn=self._conn,
+                )
+                if rows:
+                    return rows
+            except Exception as e:
+                logger.warning(f"ngram 倒排回退失败，将按配置决定是否使用 LIKE 回退: {e}")
+
+        if not self.config.enable_like_fallback:
+            return []
+
+        conditions = " OR ".join(["p.content LIKE ?"] * len(uniq_tokens))
+        params: List[Any] = [f"%{tok}%" for tok in uniq_tokens]
+        scan_limit = max(int(limit) * 8, 200)
+        params.append(scan_limit)
+
+        sql = f"""
+            SELECT p.hash, p.content
+            FROM paragraphs p
+            WHERE (p.is_deleted IS NULL OR p.is_deleted = 0)
+              AND ({conditions})
+            LIMIT ?
+        """
+        rows = self.metadata_store.query(sql, tuple(params))
+        if not rows:
+            return []
+
+        scored: List[Dict[str, Any]] = []
+        token_count = max(1, len(uniq_tokens))
+        for row in rows:
+            content = str(row.get("content") or "")
+            content_low = content.lower()
+            matched = [tok for tok in uniq_tokens if tok in content_low]
+            if not matched:
+                continue
+            coverage = len(matched) / token_count
+            length_bonus = sum(len(tok) for tok in matched) / max(1, len(content_low))
+            # 兜底路径使用相对分，保持与上层接口兼容。
+            fallback_score = coverage * 0.8 + length_bonus * 0.2
+            scored.append(
+                {
+                    "hash": row["hash"],
+                    "content": content[: self.config.max_doc_len] if self.config.max_doc_len > 0 else content,
+                    "bm25_score": -float(fallback_score),
+                    "fallback_score": float(fallback_score),
+                }
+            )
+
+        scored.sort(key=lambda x: x["fallback_score"], reverse=True)
+        return scored[:limit]
+
+    def search(self, query: str, k: int = 20) -> List[Dict[str, Any]]:
+        """执行 BM25 检索。"""
+        if not self.config.enabled:
+            return []
+        if self.config.lazy_load and not self.loaded:
+            if not self.ensure_loaded():
+                return []
+        if not self.loaded:
+            return []
+        # 关系稀疏检索可独立开关，运行时开启后也能按需补齐 schema/回填。
+        self.metadata_store.ensure_relations_fts_schema(conn=self._conn)
+        self.metadata_store.ensure_relations_fts_backfilled(conn=self._conn)
+
+        tokens = self._tokenize(query)
+        match_query = self._build_match_query(tokens)
+        if not match_query:
+            return []
+
+        limit = max(1, int(k))
+        rows = self.metadata_store.fts_search_bm25(
+            match_query=match_query,
+            limit=limit,
+            max_doc_len=self.config.max_doc_len,
+            conn=self._conn,
+        )
+        if not rows:
+            rows = self._fallback_substring_search(tokens=tokens, limit=limit)
+
+        results: List[Dict[str, Any]] = []
+        for rank, row in enumerate(rows, start=1):
+            bm25_score = float(row.get("bm25_score", 0.0))
+            results.append(
+                {
+                    "hash": row["hash"],
+                    "content": row["content"],
+                    "rank": rank,
+                    "bm25_score": bm25_score,
+                    "score": -bm25_score,  # bm25 越小越相关，这里取反作为相对分数
+                }
+            )
+        return results
+
+    def search_relations(self, query: str, k: int = 20) -> List[Dict[str, Any]]:
+        """执行关系稀疏检索（FTS5 + BM25）。"""
+        if not self.config.enabled or not self.config.enable_relation_sparse_fallback:
+            return []
+        if self.config.lazy_load and not self.loaded:
+            if not self.ensure_loaded():
+                return []
+        if not self.loaded:
+            return []
+
+        tokens = self._tokenize(query)
+        match_query = self._build_match_query(tokens)
+        if not match_query:
+            return []
+
+        rows = self.metadata_store.fts_search_relations_bm25(
+            match_query=match_query,
+            limit=max(1, int(k)),
+            max_doc_len=self.config.relation_max_doc_len,
+            conn=self._conn,
+        )
+        out: List[Dict[str, Any]] = []
+        for rank, row in enumerate(rows, start=1):
+            bm25_score = float(row.get("bm25_score", 0.0))
+            out.append(
+                {
+                    "hash": row["hash"],
+                    "subject": row["subject"],
+                    "predicate": row["predicate"],
+                    "object": row["object"],
+                    "content": row["content"],
+                    "rank": rank,
+                    "bm25_score": bm25_score,
+                    "score": -bm25_score,
+                }
+            )
+        return out
+
+    def upsert_paragraph(self, paragraph_hash: str) -> bool:
+        if not self.loaded:
+            return False
+        return self.metadata_store.fts_upsert_paragraph(paragraph_hash, conn=self._conn)
+
+    def delete_paragraph(self, paragraph_hash: str) -> bool:
+        if not self.loaded:
+            return False
+        return self.metadata_store.fts_delete_paragraph(paragraph_hash, conn=self._conn)
+
+    def unload(self) -> None:
+        """卸载 BM25 连接并尽量释放内存。"""
+        if self._conn is not None:
+            try:
+                if self.config.shrink_memory_on_unload:
+                    self.metadata_store.shrink_memory(conn=self._conn)
+            except Exception:
+                pass
+            try:
+                self._conn.close()
+            except Exception:
+                pass
+        self._conn = None
+        self._loaded = False
+        logger.info("SparseBM25Index unloaded")
+
+    def stats(self) -> Dict[str, Any]:
+        doc_count = 0
+        if self.loaded:
+            doc_count = self.metadata_store.fts_doc_count(conn=self._conn)
+        return {
+            "enabled": self.config.enabled,
+            "backend": self.config.backend,
+            "mode": self.config.mode,
+            "tokenizer_mode": self.config.tokenizer_mode,
+            "enable_ngram_fallback_index": self.config.enable_ngram_fallback_index,
+            "enable_like_fallback": self.config.enable_like_fallback,
+            "enable_relation_sparse_fallback": self.config.enable_relation_sparse_fallback,
+            "loaded": self.loaded,
+            "has_jieba": HAS_JIEBA,
+            "doc_count": doc_count,
+        }

plugins/A_memorix/core/retrieval/threshold.py

@@ -0,0 +1,450 @@
+"""
+动态阈值过滤器
+
+根据检索结果的分布特征自适应调整过滤阈值。
+"""
+
+import numpy as np
+from typing import List, Dict, Any, Optional, Tuple, Union
+from dataclasses import dataclass
+from enum import Enum
+
+from src.common.logger import get_logger
+from .dual_path import RetrievalResult
+
+logger = get_logger("A_Memorix.DynamicThresholdFilter")
+
+
+class ThresholdMethod(Enum):
+    """阈值计算方法"""
+
+    PERCENTILE = "percentile"  # 百分位数
+    STD_DEV = "std_dev"        # 标准差
+    GAP_DETECTION = "gap_detection"  # 跳变检测
+    ADAPTIVE = "adaptive"      # 自适应（综合多种方法）
+
+
+@dataclass
+class ThresholdConfig:
+    """
+    阈值配置
+
+    属性：
+        method: 阈值计算方法
+        min_threshold: 最小阈值（绝对值）
+        max_threshold: 最大阈值（绝对值）
+        percentile: 百分位数（用于percentile方法）
+        std_multiplier: 标准差倍数（用于std_dev方法）
+        min_results: 最少保留结果数
+        enable_auto_adjust: 是否自动调整参数
+    """
+
+    method: ThresholdMethod = ThresholdMethod.ADAPTIVE
+    min_threshold: float = 0.3
+    max_threshold: float = 0.95
+    percentile: float = 75.0  # 百分位数
+    std_multiplier: float = 1.5  # 标准差倍数
+    min_results: int = 3  # 最少保留结果数
+    enable_auto_adjust: bool = True
+
+    def __post_init__(self):
+        """验证配置"""
+        if not 0 <= self.min_threshold <= 1:
+            raise ValueError(f"min_threshold必须在[0, 1]之间: {self.min_threshold}")
+
+        if not 0 <= self.max_threshold <= 1:
+            raise ValueError(f"max_threshold必须在[0, 1]之间: {self.max_threshold}")
+
+        if self.min_threshold >= self.max_threshold:
+            raise ValueError(f"min_threshold必须小于max_threshold")
+
+        if not 0 <= self.percentile <= 100:
+            raise ValueError(f"percentile必须在[0, 100]之间: {self.percentile}")
+
+        if self.std_multiplier <= 0:
+            raise ValueError(f"std_multiplier必须大于0: {self.std_multiplier}")
+
+        if self.min_results < 0:
+            raise ValueError(f"min_results必须大于等于0: {self.min_results}")
+
+
+class DynamicThresholdFilter:
+    """
+    动态阈值过滤器
+
+    功能：
+    - 基于结果分布自适应计算阈值
+    - 多种阈值计算方法
+    - 自动参数调整
+    - 统计信息收集
+
+    参数：
+        config: 阈值配置
+    """
+
+    def __init__(
+        self,
+        config: Optional[ThresholdConfig] = None,
+    ):
+        """
+        初始化动态阈值过滤器
+
+        Args:
+            config: 阈值配置
+        """
+        self.config = config or ThresholdConfig()
+
+        # 统计信息
+        self._total_filtered = 0
+        self._total_processed = 0
+        self._threshold_history: List[float] = []
+
+        logger.info(
+            f"DynamicThresholdFilter 初始化: "
+            f"method={self.config.method.value}, "
+            f"min_threshold={self.config.min_threshold}"
+        )
+
+    def filter(
+        self,
+        results: List[RetrievalResult],
+        return_threshold: bool = False,
+    ) -> Union[List[RetrievalResult], Tuple[List[RetrievalResult], float]]:
+        """
+        过滤检索结果
+
+        Args:
+            results: 检索结果列表
+            return_threshold: 是否返回使用的阈值
+
+        Returns:
+            过滤后的结果列表，或 (结果列表, 阈值) 元组
+        """
+        if not results:
+            logger.debug("结果列表为空，无需过滤")
+            return ([], 0.0) if return_threshold else []
+
+        self._total_processed += len(results)
+
+        # 提取分数
+        scores = np.array([r.score for r in results])
+
+        # 计算阈值
+        threshold = self._compute_threshold(scores, results)
+
+        # 记录阈值
+        self._threshold_history.append(threshold)
+
+        # 应用阈值过滤
+        filtered_results = [
+            r for r in results
+            if r.score >= threshold
+        ]
+
+        # 确保至少保留min_results个结果
+        if len(filtered_results) < self.config.min_results:
+            # 按分数排序，取前min_results个
+            sorted_results = sorted(results, key=lambda x: x.score, reverse=True)
+            filtered_results = sorted_results[:self.config.min_results]
+            threshold = filtered_results[-1].score if filtered_results else 0.0
+
+        self._total_filtered += len(results) - len(filtered_results)
+
+        logger.info(
+            f"过滤完成: {len(results)} -> {len(filtered_results)} "
+            f"(threshold={threshold:.3f})"
+        )
+
+        if return_threshold:
+            return filtered_results, threshold
+        return filtered_results
+
+    def _compute_threshold(
+        self,
+        scores: np.ndarray,
+        results: List[RetrievalResult],
+    ) -> float:
+        """
+        计算阈值
+
+        Args:
+            scores: 分数数组
+            results: 检索结果列表
+
+        Returns:
+            阈值
+        """
+        if self.config.method == ThresholdMethod.PERCENTILE:
+            threshold = self._percentile_threshold(scores)
+        elif self.config.method == ThresholdMethod.STD_DEV:
+            threshold = self._std_dev_threshold(scores)
+        elif self.config.method == ThresholdMethod.GAP_DETECTION:
+            threshold = self._gap_detection_threshold(scores)
+        else:  # ADAPTIVE
+            # 自适应方法：综合多种方法
+            thresholds = [
+                self._percentile_threshold(scores),
+                self._std_dev_threshold(scores),
+                self._gap_detection_threshold(scores),
+            ]
+            # 使用中位数作为最终阈值
+            threshold = float(np.median(thresholds))
+
+        # 限制在[min_threshold, max_threshold]范围内
+        threshold = np.clip(
+            threshold,
+            self.config.min_threshold,
+            self.config.max_threshold,
+        )
+
+        # 自动调整
+        if self.config.enable_auto_adjust:
+            threshold = self._auto_adjust_threshold(threshold, scores)
+
+        return float(threshold)
+
+    def _percentile_threshold(self, scores: np.ndarray) -> float:
+        """
+        基于百分位数计算阈值
+
+        Args:
+            scores: 分数数组
+
+        Returns:
+            阈值
+        """
+        percentile = self.config.percentile
+        threshold = float(np.percentile(scores, percentile))
+
+        logger.debug(f"百分位数阈值: {threshold:.3f} (percentile={percentile})")
+        return threshold
+
+    def _std_dev_threshold(self, scores: np.ndarray) -> float:
+        """
+        基于标准差计算阈值
+
+        threshold = mean - std_multiplier * std
+
+        Args:
+            scores: 分数数组
+
+        Returns:
+            阈值
+        """
+        mean = float(np.mean(scores))
+        std = float(np.std(scores))
+        multiplier = self.config.std_multiplier
+
+        threshold = mean - multiplier * std
+
+        logger.debug(f"标准差阈值: {threshold:.3f} (mean={mean:.3f}, std={std:.3f})")
+        return threshold
+
+    def _gap_detection_threshold(self, scores: np.ndarray) -> float:
+        """
+        基于跳变检测计算阈值
+
+        找到分数分布中最大的"跳变"位置，以此为阈值
+
+        Args:
+            scores: 分数数组（降序排列）
+
+        Returns:
+            阈值
+        """
+        # 降序排列
+        sorted_scores = np.sort(scores)[::-1]
+
+        if len(sorted_scores) < 2:
+            return float(sorted_scores[0]) if len(sorted_scores) > 0 else 0.0
+
+        # 计算相邻分数的差值
+        gaps = np.diff(sorted_scores)
+
+        # 找到最大的跳变位置
+        max_gap_idx = int(np.argmax(gaps))
+
+        # 阈值为跳变后的分数
+        threshold = float(sorted_scores[max_gap_idx + 1])
+
+        logger.debug(
+            f"跳变检测阈值: {threshold:.3f} "
+            f"(gap={gaps[max_gap_idx]:.3f}, idx={max_gap_idx})"
+        )
+        return threshold
+
+    def _auto_adjust_threshold(
+        self,
+        threshold: float,
+        scores: np.ndarray,
+    ) -> float:
+        """
+        自动调整阈值
+
+        基于历史阈值和当前分数分布调整
+
+        Args:
+            threshold: 当前阈值
+            scores: 分数数组
+
+        Returns:
+            调整后的阈值
+        """
+        if not self._threshold_history:
+            return threshold
+
+        # 计算历史阈值的移动平均
+        recent_thresholds = self._threshold_history[-10:]  # 最近10次
+        avg_threshold = float(np.mean(recent_thresholds))
+
+        # 当前阈值与历史平均的差异
+        diff = threshold - avg_threshold
+
+        # 如果差异过大（>0.2），向历史平均靠拢
+        if abs(diff) > 0.2:
+            adjusted_threshold = avg_threshold + diff * 0.5  # 向中间靠拢50%
+            logger.debug(
+                f"阈值调整: {threshold:.3f} -> {adjusted_threshold:.3f} "
+                f"(历史平均={avg_threshold:.3f})"
+            )
+            return adjusted_threshold
+
+        return threshold
+
+    def filter_by_confidence(
+        self,
+        results: List[RetrievalResult],
+        min_confidence: float = 0.5,
+    ) -> List[RetrievalResult]:
+        """
+        基于置信度过滤结果
+
+        Args:
+            results: 检索结果列表
+            min_confidence: 最小置信度
+
+        Returns:
+            过滤后的结果列表
+        """
+        filtered = []
+        for result in results:
+            # 对于关系结果，使用confidence字段
+            if result.result_type == "relation":
+                confidence = result.metadata.get("confidence", 1.0)
+                if confidence >= min_confidence:
+                    filtered.append(result)
+            else:
+                # 对于段落结果，直接使用分数
+                if result.score >= min_confidence:
+                    filtered.append(result)
+
+        logger.info(
+            f"置信度过滤: {len(results)} -> {len(filtered)} "
+            f"(min_confidence={min_confidence})"
+        )
+
+        return filtered
+
+    def filter_by_diversity(
+        self,
+        results: List[RetrievalResult],
+        similarity_threshold: float = 0.9,
+        top_k: int = 10,
+    ) -> List[RetrievalResult]:
+        """
+        基于多样性过滤结果（去除重复）
+
+        Args:
+            results: 检索结果列表
+            similarity_threshold: 相似度阈值（高于此值视为重复）
+            top_k: 最多保留结果数
+
+        Returns:
+            过滤后的结果列表
+        """
+        if not results:
+            return []
+
+        # 按分数排序
+        sorted_results = sorted(results, key=lambda x: x.score, reverse=True)
+
+        # 贪心选择：选择与已选结果相似度低的结果
+        selected = []
+        selected_hashes = []
+
+        for result in sorted_results:
+            if len(selected) >= top_k:
+                break
+
+            # 检查与已选结果的相似度
+            is_duplicate = False
+            for selected_hash in selected_hashes:
+                # 简单判断：基于hash的前缀
+                if result.hash_value[:8] == selected_hash[:8]:
+                    is_duplicate = True
+                    break
+
+            if not is_duplicate:
+                selected.append(result)
+                selected_hashes.append(result.hash_value)
+
+        logger.info(
+            f"多样性过滤: {len(results)} -> {len(selected)} "
+            f"(similarity_threshold={similarity_threshold})"
+        )
+
+        return selected
+
+    def get_statistics(self) -> Dict[str, Any]:
+        """
+        获取统计信息
+
+        Returns:
+            统计信息字典
+        """
+        filter_rate = (
+            self._total_filtered / self._total_processed
+            if self._total_processed > 0
+            else 0.0
+        )
+
+        stats = {
+            "config": {
+                "method": self.config.method.value,
+                "min_threshold": self.config.min_threshold,
+                "max_threshold": self.config.max_threshold,
+                "percentile": self.config.percentile,
+                "std_multiplier": self.config.std_multiplier,
+                "min_results": self.config.min_results,
+                "enable_auto_adjust": self.config.enable_auto_adjust,
+            },
+            "statistics": {
+                "total_processed": self._total_processed,
+                "total_filtered": self._total_filtered,
+                "filter_rate": filter_rate,
+                "avg_threshold": float(np.mean(self._threshold_history))
+                if self._threshold_history else 0.0,
+                "threshold_count": len(self._threshold_history),
+            },
+        }
+
+        if self._threshold_history:
+            stats["statistics"]["min_threshold_used"] = float(np.min(self._threshold_history))
+            stats["statistics"]["max_threshold_used"] = float(np.max(self._threshold_history))
+
+        return stats
+
+    def reset_statistics(self) -> None:
+        """重置统计信息"""
+        self._total_filtered = 0
+        self._total_processed = 0
+        self._threshold_history.clear()
+        logger.info("统计信息已重置")
+
+    def __repr__(self) -> str:
+        return (
+            f"DynamicThresholdFilter("
+            f"method={self.config.method.value}, "
+            f"min_threshold={self.config.min_threshold}, "
+            f"filtered={self._total_filtered}/{self._total_processed})"
+        )

plugins/A_memorix/core/runtime/__init__.py

@@ -0,0 +1,8 @@
+"""SDK runtime exports for A_Memorix."""
+
+from .sdk_memory_kernel import KernelSearchRequest, SDKMemoryKernel
+
+__all__ = [
+    "KernelSearchRequest",
+    "SDKMemoryKernel",
+]

plugins/A_memorix/core/runtime/sdk_memory_kernel.py

@@ -0,0 +1,579 @@
+from __future__ import annotations
+
+import time
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any, Dict, Iterable, List, Optional, Sequence
+
+from src.common.logger import get_logger
+
+from ..embedding import create_embedding_api_adapter
+from ..retrieval import (
+    DualPathRetriever,
+    DualPathRetrieverConfig,
+    RetrievalResult,
+    SparseBM25Config,
+    SparseBM25Index,
+    TemporalQueryOptions,
+)
+from ..storage import GraphStore, MetadataStore, QuantizationType, SparseMatrixFormat, VectorStore
+from ..utils.aggregate_query_service import AggregateQueryService
+from ..utils.episode_retrieval_service import EpisodeRetrievalService
+from ..utils.hash import normalize_text
+from ..utils.relation_write_service import RelationWriteService
+
+logger = get_logger("A_Memorix.SDKMemoryKernel")
+
+
+@dataclass
+class KernelSearchRequest:
+    query: str = ""
+    limit: int = 5
+    mode: str = "hybrid"
+    chat_id: str = ""
+    person_id: str = ""
+    time_start: Optional[float] = None
+    time_end: Optional[float] = None
+
+
+class SDKMemoryKernel:
+    def __init__(self, *, plugin_root: Path, config: Optional[Dict[str, Any]] = None) -> None:
+        self.plugin_root = Path(plugin_root).resolve()
+        self.config = config or {}
+        storage_cfg = self._cfg("storage", {}) or {}
+        data_dir = str(storage_cfg.get("data_dir", "./data") or "./data")
+        self.data_dir = (self.plugin_root / data_dir).resolve() if data_dir.startswith(".") else Path(data_dir)
+        self.embedding_dimension = max(32, int(self._cfg("embedding.dimension", 256)))
+        self.relation_vectors_enabled = bool(self._cfg("retrieval.relation_vectorization.enabled", False))
+
+        self.embedding_manager = None
+        self.vector_store: Optional[VectorStore] = None
+        self.graph_store: Optional[GraphStore] = None
+        self.metadata_store: Optional[MetadataStore] = None
+        self.relation_write_service: Optional[RelationWriteService] = None
+        self.sparse_index = None
+        self.retriever: Optional[DualPathRetriever] = None
+        self.episode_retriever: Optional[EpisodeRetrievalService] = None
+        self.aggregate_query_service: Optional[AggregateQueryService] = None
+        self._initialized = False
+        self._last_maintenance_at: Optional[float] = None
+
+    def _cfg(self, key: str, default: Any = None) -> Any:
+        current: Any = self.config
+        if key in {"storage", "embedding", "retrieval"} and isinstance(current, dict):
+            return current.get(key, default)
+        for part in key.split("."):
+            if isinstance(current, dict) and part in current:
+                current = current[part]
+            else:
+                return default
+        return current
+
+    async def initialize(self) -> None:
+        if self._initialized:
+            return
+        self.data_dir.mkdir(parents=True, exist_ok=True)
+        self.embedding_manager = create_embedding_api_adapter(
+            batch_size=int(self._cfg("embedding.batch_size", 32)),
+            max_concurrent=int(self._cfg("embedding.max_concurrent", 5)),
+            default_dimension=self.embedding_dimension,
+            model_name=str(self._cfg("embedding.model_name", "hash-v1")),
+            retry_config=self._cfg("embedding.retry", {}) or {},
+        )
+        self.embedding_dimension = int(await self.embedding_manager._detect_dimension())
+        self.vector_store = VectorStore(
+            dimension=self.embedding_dimension,
+            quantization_type=QuantizationType.INT8,
+            data_dir=self.data_dir / "vectors",
+        )
+        self.graph_store = GraphStore(matrix_format=SparseMatrixFormat.CSR, data_dir=self.data_dir / "graph")
+        self.metadata_store = MetadataStore(data_dir=self.data_dir / "metadata")
+        self.metadata_store.connect()
+        if self.vector_store.has_data():
+            self.vector_store.load()
+            self.vector_store.warmup_index(force_train=True)
+        if self.graph_store.has_data():
+            self.graph_store.load()
+
+        sparse_cfg = self._cfg("retrieval.sparse", {}) or {}
+        self.sparse_index = SparseBM25Index(metadata_store=self.metadata_store, config=SparseBM25Config(**sparse_cfg))
+        if getattr(self.sparse_index.config, "enabled", False):
+            self.sparse_index.ensure_loaded()
+
+        self.relation_write_service = RelationWriteService(
+            metadata_store=self.metadata_store,
+            graph_store=self.graph_store,
+            vector_store=self.vector_store,
+            embedding_manager=self.embedding_manager,
+        )
+        self.retriever = DualPathRetriever(
+            vector_store=self.vector_store,
+            graph_store=self.graph_store,
+            metadata_store=self.metadata_store,
+            embedding_manager=self.embedding_manager,
+            sparse_index=self.sparse_index,
+            config=DualPathRetrieverConfig(
+                top_k_paragraphs=int(self._cfg("retrieval.top_k_paragraphs", 24)),
+                top_k_relations=int(self._cfg("retrieval.top_k_relations", 12)),
+                top_k_final=int(self._cfg("retrieval.top_k_final", 10)),
+                alpha=float(self._cfg("retrieval.alpha", 0.5)),
+                enable_ppr=bool(self._cfg("retrieval.enable_ppr", True)),
+                ppr_alpha=float(self._cfg("retrieval.ppr_alpha", 0.85)),
+                ppr_concurrency_limit=int(self._cfg("retrieval.ppr_concurrency_limit", 4)),
+                enable_parallel=bool(self._cfg("retrieval.enable_parallel", True)),
+                sparse=sparse_cfg,
+                fusion=self._cfg("retrieval.fusion", {}) or {},
+                graph_recall=self._cfg("retrieval.search.graph_recall", {}) or {},
+                relation_intent=self._cfg("retrieval.search.relation_intent", {}) or {},
+            ),
+        )
+        self.episode_retriever = EpisodeRetrievalService(metadata_store=self.metadata_store, retriever=self.retriever)
+        self.aggregate_query_service = AggregateQueryService(plugin_config=self.config)
+        self._initialized = True
+
+    def close(self) -> None:
+        if self.vector_store is not None:
+            self.vector_store.save()
+        if self.graph_store is not None:
+            self.graph_store.save()
+        if self.metadata_store is not None:
+            self.metadata_store.close()
+        self._initialized = False
+
+    async def ingest_summary(
+        self,
+        *,
+        external_id: str,
+        chat_id: str,
+        text: str,
+        participants: Optional[Sequence[str]] = None,
+        time_start: Optional[float] = None,
+        time_end: Optional[float] = None,
+        tags: Optional[Sequence[str]] = None,
+        metadata: Optional[Dict[str, Any]] = None,
+    ) -> Dict[str, Any]:
+        summary_meta = dict(metadata or {})
+        summary_meta.setdefault("kind", "chat_summary")
+        return await self.ingest_text(
+            external_id=external_id,
+            source_type="chat_summary",
+            text=text,
+            chat_id=chat_id,
+            participants=participants,
+            time_start=time_start,
+            time_end=time_end,
+            tags=tags,
+            metadata=summary_meta,
+        )
+
+    async def ingest_text(
+        self,
+        *,
+        external_id: str,
+        source_type: str,
+        text: str,
+        chat_id: str = "",
+        person_ids: Optional[Sequence[str]] = None,
+        participants: Optional[Sequence[str]] = None,
+        timestamp: Optional[float] = None,
+        time_start: Optional[float] = None,
+        time_end: Optional[float] = None,
+        tags: Optional[Sequence[str]] = None,
+        metadata: Optional[Dict[str, Any]] = None,
+        entities: Optional[Sequence[str]] = None,
+        relations: Optional[Sequence[Dict[str, Any]]] = None,
+    ) -> Dict[str, Any]:
+        await self.initialize()
+        assert self.metadata_store and self.vector_store and self.graph_store and self.embedding_manager
+        assert self.relation_write_service
+        content = normalize_text(text)
+        if not content:
+            return {"stored_ids": [], "skipped_ids": [external_id], "reason": "empty_text"}
+        if ref := self.metadata_store.get_external_memory_ref(external_id):
+            return {"stored_ids": [], "skipped_ids": [str(ref.get("paragraph_hash", ""))], "reason": "exists"}
+
+        person_tokens = self._tokens(person_ids)
+        participant_tokens = self._tokens(participants)
+        entity_tokens = self._merge_tokens(entities, person_tokens, participant_tokens)
+        source = self._build_source(source_type, chat_id, person_tokens)
+        paragraph_meta = dict(metadata or {})
+        paragraph_meta.update(
+            {
+                "external_id": external_id,
+                "source_type": str(source_type or "").strip(),
+                "chat_id": str(chat_id or "").strip(),
+                "person_ids": person_tokens,
+                "participants": participant_tokens,
+                "tags": self._tokens(tags),
+            }
+        )
+        paragraph_hash = self.metadata_store.add_paragraph(
+            content=content,
+            source=source,
+            metadata=paragraph_meta,
+            knowledge_type="factual" if source_type == "person_fact" else "narrative" if source_type == "chat_summary" else "mixed",
+            time_meta=self._time_meta(timestamp, time_start, time_end),
+        )
+        embedding = await self.embedding_manager.encode(content)
+        self.vector_store.add(vectors=embedding.reshape(1, -1), ids=[paragraph_hash])
+        for name in entity_tokens:
+            self.metadata_store.add_entity(name=name, source_paragraph=paragraph_hash)
+
+        stored_relations: List[str] = []
+        for row in [dict(item) for item in (relations or []) if isinstance(item, dict)]:
+            s = str(row.get("subject", "") or "").strip()
+            p = str(row.get("predicate", "") or "").strip()
+            o = str(row.get("object", "") or "").strip()
+            if not (s and p and o):
+                continue
+            result = await self.relation_write_service.upsert_relation_with_vector(
+                subject=s,
+                predicate=p,
+                obj=o,
+                confidence=float(row.get("confidence", 1.0) or 1.0),
+                source_paragraph=paragraph_hash,
+                metadata={"external_id": external_id, "source_type": source_type},
+                write_vector=self.relation_vectors_enabled,
+            )
+            self.metadata_store.link_paragraph_relation(paragraph_hash, result.hash_value)
+            stored_relations.append(result.hash_value)
+
+        self.metadata_store.upsert_external_memory_ref(
+            external_id=external_id,
+            paragraph_hash=paragraph_hash,
+            source_type=source_type,
+            metadata={"chat_id": chat_id, "person_ids": person_tokens},
+        )
+        self._persist()
+        self.rebuild_episodes_for_sources([source])
+        for person_id in person_tokens:
+            await self.refresh_person_profile(person_id)
+        return {"stored_ids": [paragraph_hash, *stored_relations], "skipped_ids": []}
+
+    async def search_memory(self, request: KernelSearchRequest) -> Dict[str, Any]:
+        await self.initialize()
+        assert self.retriever and self.episode_retriever and self.aggregate_query_service
+        mode = str(request.mode or "hybrid").strip().lower() or "hybrid"
+        clean_query = str(request.query or "").strip()
+        limit = max(1, int(request.limit or 5))
+        temporal = self._temporal(request)
+        if mode == "episode":
+            rows = await self.episode_retriever.query(
+                query=clean_query,
+                top_k=limit,
+                time_from=request.time_start,
+                time_to=request.time_end,
+                source=self._chat_source(request.chat_id),
+            )
+            hits = [self._episode_hit(row) for row in rows]
+            return {"summary": self._summary(hits), "hits": hits}
+        if mode == "aggregate":
+            payload = await self.aggregate_query_service.execute(
+                query=clean_query,
+                top_k=limit,
+                mix=True,
+                mix_top_k=limit,
+                time_from=str(request.time_start) if request.time_start is not None else None,
+                time_to=str(request.time_end) if request.time_end is not None else None,
+                search_runner=lambda: self._aggregate_search(clean_query, limit, temporal),
+                time_runner=lambda: self._aggregate_time(clean_query, limit, temporal),
+                episode_runner=lambda: self._aggregate_episode(clean_query, limit, request),
+            )
+            hits = [dict(item) for item in payload.get("mixed_results", []) if isinstance(item, dict)]
+            for item in hits:
+                item.setdefault("metadata", {})
+            return {"summary": self._summary(hits), "hits": hits}
+        results = await self.retriever.retrieve(query=clean_query, top_k=limit, temporal=temporal)
+        hits = [self._retrieval_hit(item) for item in results]
+        return {"summary": self._summary(self._filter_hits(hits, request.person_id)), "hits": self._filter_hits(hits, request.person_id)}
+
+    async def get_person_profile(self, *, person_id: str, chat_id: str = "", limit: int = 10) -> Dict[str, Any]:
+        _ = chat_id
+        await self.initialize()
+        assert self.metadata_store
+        snapshot = self.metadata_store.get_latest_person_profile_snapshot(person_id) or await self.refresh_person_profile(person_id, limit=limit)
+        evidence = []
+        for hash_value in snapshot.get("evidence_ids", [])[: max(1, int(limit))]:
+            paragraph = self.metadata_store.get_paragraph(hash_value)
+            if paragraph is not None:
+                evidence.append({"hash": hash_value, "content": str(paragraph.get("content", "") or "")[:220], "metadata": paragraph.get("metadata", {}) or {}})
+        text = str(snapshot.get("profile_text", "") or "").strip()
+        traits = [line.strip("- ").strip() for line in text.splitlines() if line.strip()][:8]
+        return {"summary": text, "traits": traits, "evidence": evidence}
+
+    async def refresh_person_profile(self, person_id: str, limit: int = 10) -> Dict[str, Any]:
+        await self.initialize()
+        assert self.metadata_store
+        rows = self.metadata_store.query(
+            """
+            SELECT DISTINCT p.*
+            FROM paragraphs p
+            JOIN paragraph_entities pe ON pe.paragraph_hash = p.hash
+            JOIN entities e ON e.hash = pe.entity_hash
+            WHERE e.name = ?
+              AND (p.is_deleted IS NULL OR p.is_deleted = 0)
+            ORDER BY COALESCE(p.event_time_end, p.event_time_start, p.event_time, p.updated_at, p.created_at) DESC
+            LIMIT ?
+            """,
+            (person_id, max(1, int(limit)) * 3),
+        )
+        evidence_ids = [str(row.get("hash", "") or "") for row in rows if str(row.get("hash", "")).strip()]
+        vector_evidence = [{"hash": str(row.get("hash", "") or ""), "type": "paragraph", "score": 0.0, "content": str(row.get("content", "") or "")[:220], "metadata": row.get("metadata", {}) or {}} for row in rows[: max(1, int(limit))]]
+        relation_edges = [{"hash": str(row.get("hash", "") or ""), "subject": str(row.get("subject", "") or ""), "predicate": str(row.get("predicate", "") or ""), "object": str(row.get("object", "") or ""), "confidence": float(row.get("confidence", 1.0) or 1.0)} for row in self.metadata_store.get_relations(subject=person_id)[:limit]]
+        if relation_edges:
+            profile_text = "\n".join(f"{item['subject']} {item['predicate']} {item['object']}" for item in relation_edges[:6])
+        elif vector_evidence:
+            profile_text = "\n".join(f"- {item['content']}" for item in vector_evidence[:6])
+        else:
+            profile_text = "暂无稳定画像证据。"
+        return self.metadata_store.upsert_person_profile_snapshot(
+            person_id=person_id,
+            profile_text=profile_text,
+            aliases=[person_id],
+            relation_edges=relation_edges,
+            vector_evidence=vector_evidence,
+            evidence_ids=evidence_ids[: max(1, int(limit))],
+            expires_at=time.time() + 6 * 3600,
+            source_note="sdk_memory_kernel",
+        )
+
+    async def maintain_memory(self, *, action: str, target: str, hours: Optional[float] = None, reason: str = "") -> Dict[str, Any]:
+        _ = reason
+        await self.initialize()
+        assert self.metadata_store
+        hashes = self._resolve_relation_hashes(target)
+        if not hashes:
+            return {"success": False, "detail": "未命中可维护关系"}
+        act = str(action or "").strip().lower()
+        if act == "reinforce":
+            self.metadata_store.reinforce_relations(hashes)
+        elif act == "protect":
+            ttl_seconds = max(0.0, float(hours or 0.0)) * 3600.0
+            self.metadata_store.protect_relations(hashes, ttl_seconds=ttl_seconds, is_pinned=ttl_seconds <= 0)
+        elif act == "restore":
+            restored = sum(1 for hash_value in hashes if self.metadata_store.restore_relation(hash_value))
+            if restored <= 0:
+                return {"success": False, "detail": "未恢复任何关系"}
+        else:
+            return {"success": False, "detail": f"不支持的维护动作: {act}"}
+        self._last_maintenance_at = time.time()
+        self._persist()
+        return {"success": True, "detail": f"{act} {len(hashes)} 条关系"}
+
+    def rebuild_episodes_for_sources(self, sources: Iterable[str]) -> int:
+        assert self.metadata_store
+        rebuilt = 0
+        for source in self._tokens(sources):
+            rows = self.metadata_store.query(
+                """
+                SELECT * FROM paragraphs
+                WHERE source = ?
+                  AND (is_deleted IS NULL OR is_deleted = 0)
+                ORDER BY COALESCE(event_time_start, event_time, created_at) ASC, hash ASC
+                """,
+                (source,),
+            )
+            if not rows:
+                continue
+            paragraph_hashes = [str(row.get("hash", "") or "") for row in rows if str(row.get("hash", "")).strip()]
+            payload = self.metadata_store.upsert_episode(
+                {
+                    "source": source,
+                    "title": str((rows[0].get("metadata", {}) or {}).get("theme", "") or f"{source} 情景记忆")[:80],
+                    "summary": "；".join(str(row.get("content", "") or "").strip().replace("\n", " ")[:120] for row in rows[:3] if str(row.get("content", "") or "").strip())[:500] or "自动构建的情景记忆。",
+                    "participants": self._episode_participants(rows),
+                    "keywords": self._episode_keywords(rows),
+                    "evidence_ids": paragraph_hashes,
+                    "paragraph_count": len(paragraph_hashes),
+                    "event_time_start": self._time_bound(rows, "event_time_start", "event_time", reverse=False),
+                    "event_time_end": self._time_bound(rows, "event_time_end", "event_time", reverse=True),
+                    "time_granularity": "day",
+                    "time_confidence": 0.7,
+                    "llm_confidence": 0.0,
+                    "segmentation_model": "rule_based_sdk",
+                    "segmentation_version": "1",
+                }
+            )
+            self.metadata_store.bind_episode_paragraphs(payload["episode_id"], paragraph_hashes)
+            rebuilt += 1
+        return rebuilt
+
+    def memory_stats(self) -> Dict[str, Any]:
+        assert self.metadata_store
+        stats = self.metadata_store.get_statistics()
+        episodes = self.metadata_store.query("SELECT COUNT(*) AS c FROM episodes")[0]["c"]
+        profiles = self.metadata_store.query("SELECT COUNT(*) AS c FROM person_profile_snapshots")[0]["c"]
+        return {"paragraphs": int(stats.get("paragraph_count", 0) or 0), "relations": int(stats.get("relation_count", 0) or 0), "episodes": int(episodes or 0), "profiles": int(profiles or 0), "last_maintenance_at": self._last_maintenance_at}
+
+    async def _aggregate_search(self, query: str, limit: int, temporal: Optional[TemporalQueryOptions]) -> Dict[str, Any]:
+        assert self.retriever
+        hits = [self._retrieval_hit(item) for item in await self.retriever.retrieve(query=query, top_k=limit, temporal=temporal)]
+        return {"success": True, "results": hits, "count": len(hits), "query_type": "search"}
+
+    async def _aggregate_time(self, query: str, limit: int, temporal: Optional[TemporalQueryOptions]) -> Dict[str, Any]:
+        if temporal is None:
+            return {"success": False, "error": "missing temporal window", "results": []}
+        assert self.retriever
+        hits = [self._retrieval_hit(item) for item in await self.retriever.retrieve(query=query, top_k=limit, temporal=temporal)]
+        return {"success": True, "results": hits, "count": len(hits), "query_type": "time"}
+
+    async def _aggregate_episode(self, query: str, limit: int, request: KernelSearchRequest) -> Dict[str, Any]:
+        assert self.episode_retriever
+        rows = await self.episode_retriever.query(query=query, top_k=limit, time_from=request.time_start, time_to=request.time_end, source=self._chat_source(request.chat_id))
+        hits = [self._episode_hit(row) for row in rows]
+        return {"success": True, "results": hits, "count": len(hits), "query_type": "episode"}
+
+    def _persist(self) -> None:
+        if self.vector_store is not None:
+            self.vector_store.save()
+        if self.graph_store is not None:
+            self.graph_store.save()
+        if self.sparse_index is not None and getattr(self.sparse_index.config, "enabled", False):
+            self.sparse_index.ensure_loaded()
+
+    @staticmethod
+    def _tokens(values: Optional[Iterable[Any]]) -> List[str]:
+        result: List[str] = []
+        seen = set()
+        for item in values or []:
+            token = str(item or "").strip()
+            if not token or token in seen:
+                continue
+            seen.add(token)
+            result.append(token)
+        return result
+
+    @classmethod
+    def _merge_tokens(cls, *groups: Optional[Iterable[Any]]) -> List[str]:
+        merged: List[str] = []
+        seen = set()
+        for group in groups:
+            for item in cls._tokens(group):
+                if item in seen:
+                    continue
+                seen.add(item)
+                merged.append(item)
+        return merged
+
+    @staticmethod
+    def _build_source(source_type: str, chat_id: str, person_ids: Sequence[str]) -> str:
+        clean_type = str(source_type or "").strip() or "memory"
+        if clean_type == "chat_summary" and chat_id:
+            return f"chat_summary:{chat_id}"
+        if clean_type == "person_fact" and person_ids:
+            return f"person_fact:{person_ids[0]}"
+        return f"{clean_type}:{chat_id}" if chat_id else clean_type
+
+    @staticmethod
+    def _chat_source(chat_id: str) -> Optional[str]:
+        clean = str(chat_id or "").strip()
+        return f"chat_summary:{clean}" if clean else None
+
+    @staticmethod
+    def _time_meta(timestamp: Optional[float], time_start: Optional[float], time_end: Optional[float]) -> Dict[str, Any]:
+        payload: Dict[str, Any] = {}
+        if timestamp is not None:
+            payload["event_time"] = float(timestamp)
+        if time_start is not None:
+            payload["event_time_start"] = float(time_start)
+        if time_end is not None:
+            payload["event_time_end"] = float(time_end)
+        if payload:
+            payload["time_granularity"] = "minute"
+            payload["time_confidence"] = 0.95
+        return payload
+
+    def _temporal(self, request: KernelSearchRequest) -> Optional[TemporalQueryOptions]:
+        if request.time_start is None and request.time_end is None and not request.chat_id:
+            return None
+        return TemporalQueryOptions(time_from=request.time_start, time_to=request.time_end, source=self._chat_source(request.chat_id))
+
+    @staticmethod
+    def _retrieval_hit(item: RetrievalResult) -> Dict[str, Any]:
+        payload = item.to_dict()
+        return {"hash": payload.get("hash", ""), "content": payload.get("content", ""), "score": payload.get("score", 0.0), "type": payload.get("type", ""), "source": payload.get("source", ""), "metadata": payload.get("metadata", {}) or {}}
+
+    @staticmethod
+    def _episode_hit(row: Dict[str, Any]) -> Dict[str, Any]:
+        return {"type": "episode", "episode_id": str(row.get("episode_id", "") or ""), "title": str(row.get("title", "") or ""), "content": str(row.get("summary", "") or ""), "score": float(row.get("lexical_score", 0.0) or 0.0), "source": "episode", "metadata": {"participants": row.get("participants", []) or [], "keywords": row.get("keywords", []) or [], "source": row.get("source"), "event_time_start": row.get("event_time_start"), "event_time_end": row.get("event_time_end")}}
+
+    @staticmethod
+    def _summary(hits: Sequence[Dict[str, Any]]) -> str:
+        if not hits:
+            return ""
+        lines = []
+        for index, item in enumerate(hits[:5], start=1):
+            content = str(item.get("content", "") or "").strip().replace("\n", " ")
+            lines.append(f"{index}. {(content[:120] + '...') if len(content) > 120 else content}")
+        return "\n".join(lines)
+
+    @staticmethod
+    def _filter_hits(hits: List[Dict[str, Any]], person_id: str) -> List[Dict[str, Any]]:
+        if not person_id:
+            return hits
+        filtered = []
+        for item in hits:
+            metadata = item.get("metadata", {}) or {}
+            if person_id in (metadata.get("person_ids", []) or []):
+                filtered.append(item)
+                continue
+            if person_id and person_id in str(item.get("content", "") or ""):
+                filtered.append(item)
+        return filtered or hits
+
+    @staticmethod
+    def _episode_participants(rows: Sequence[Dict[str, Any]]) -> List[str]:
+        seen = set()
+        result: List[str] = []
+        for row in rows:
+            meta = row.get("metadata", {}) or {}
+            for key in ("participants", "person_ids"):
+                for item in meta.get(key, []) or []:
+                    token = str(item or "").strip()
+                    if not token or token in seen:
+                        continue
+                    seen.add(token)
+                    result.append(token)
+        return result[:16]
+
+    @staticmethod
+    def _episode_keywords(rows: Sequence[Dict[str, Any]]) -> List[str]:
+        seen = set()
+        result: List[str] = []
+        for row in rows:
+            meta = row.get("metadata", {}) or {}
+            for item in meta.get("tags", []) or []:
+                token = str(item or "").strip()
+                if not token or token in seen:
+                    continue
+                seen.add(token)
+                result.append(token)
+        return result[:12]
+
+    @staticmethod
+    def _time_bound(rows: Sequence[Dict[str, Any]], primary: str, fallback: str, reverse: bool) -> Optional[float]:
+        values: List[float] = []
+        for row in rows:
+            for key in (primary, fallback):
+                value = row.get(key)
+                try:
+                    if value is not None:
+                        values.append(float(value))
+                        break
+                except Exception:
+                    continue
+        if not values:
+            return None
+        return max(values) if reverse else min(values)
+
+    def _resolve_relation_hashes(self, target: str) -> List[str]:
+        assert self.metadata_store
+        token = str(target or "").strip()
+        if not token:
+            return []
+        if len(token) == 64 and all(ch in "0123456789abcdef" for ch in token.lower()):
+            return [token]
+        hashes = self.metadata_store.search_relation_hashes_by_text(token, limit=10)
+        if hashes:
+            return hashes
+        return [str(row.get("hash", "") or "") for row in self.metadata_store.get_relations(subject=token)[:10] if str(row.get("hash", "")).strip()]

plugins/A_memorix/core/storage/__init__.py

@@ -0,0 +1,53 @@
+"""存储层"""
+
+from .vector_store import VectorStore, QuantizationType
+from .graph_store import GraphStore, SparseMatrixFormat
+from .metadata_store import MetadataStore
+from .knowledge_types import (
+    ImportStrategy,
+    KnowledgeType,
+    allowed_import_strategy_values,
+    allowed_knowledge_type_values,
+    get_knowledge_type_from_string,
+    get_import_strategy_from_string,
+    parse_import_strategy,
+    resolve_stored_knowledge_type,
+    should_extract_relations,
+    get_default_chunk_size,
+    get_type_display_name,
+    validate_stored_knowledge_type,
+)
+from .type_detection import (
+    detect_knowledge_type,
+    get_type_from_user_input,
+    looks_like_factual_text,
+    looks_like_quote_text,
+    looks_like_structured_text,
+    select_import_strategy,
+)
+
+__all__ = [
+    "VectorStore",
+    "GraphStore",
+    "MetadataStore",
+    "QuantizationType",
+    "SparseMatrixFormat",
+    "ImportStrategy",
+    "KnowledgeType",
+    "allowed_import_strategy_values",
+    "allowed_knowledge_type_values",
+    "get_knowledge_type_from_string",
+    "get_import_strategy_from_string",
+    "parse_import_strategy",
+    "resolve_stored_knowledge_type",
+    "should_extract_relations",
+    "get_default_chunk_size",
+    "get_type_display_name",
+    "validate_stored_knowledge_type",
+    "detect_knowledge_type",
+    "get_type_from_user_input",
+    "looks_like_factual_text",
+    "looks_like_quote_text",
+    "looks_like_structured_text",
+    "select_import_strategy",
+]

plugins/A_memorix/core/storage/knowledge_types.py

@@ -0,0 +1,183 @@
+"""Knowledge type and import strategy helpers."""
+
+from __future__ import annotations
+
+from enum import Enum
+from typing import Any, Optional
+
+
+class KnowledgeType(str, Enum):
+    """持久化到 paragraphs.knowledge_type 的合法类型。"""
+
+    STRUCTURED = "structured"
+    NARRATIVE = "narrative"
+    FACTUAL = "factual"
+    QUOTE = "quote"
+    MIXED = "mixed"
+
+
+class ImportStrategy(str, Enum):
+    """文本导入阶段的策略选择。"""
+
+    AUTO = "auto"
+    NARRATIVE = "narrative"
+    FACTUAL = "factual"
+    QUOTE = "quote"
+
+
+def allowed_knowledge_type_values() -> tuple[str, ...]:
+    return tuple(item.value for item in KnowledgeType)
+
+
+def allowed_import_strategy_values() -> tuple[str, ...]:
+    return tuple(item.value for item in ImportStrategy)
+
+
+def get_knowledge_type_from_string(type_str: Any) -> Optional[KnowledgeType]:
+    """从字符串解析合法的落库知识类型。"""
+
+    if not isinstance(type_str, str):
+        return None
+    normalized = type_str.lower().strip()
+    try:
+        return KnowledgeType(normalized)
+    except ValueError:
+        return None
+
+
+def get_import_strategy_from_string(value: Any) -> Optional[ImportStrategy]:
+    """从字符串解析文本导入策略。"""
+
+    if not isinstance(value, str):
+        return None
+    normalized = value.lower().strip()
+    try:
+        return ImportStrategy(normalized)
+    except ValueError:
+        return None
+
+
+def parse_import_strategy(value: Any, default: ImportStrategy = ImportStrategy.AUTO) -> ImportStrategy:
+    """解析 import strategy；非法值直接报错。"""
+
+    if value is None:
+        return default
+    if isinstance(value, ImportStrategy):
+        return value
+
+    normalized = str(value or "").strip().lower()
+    if not normalized:
+        return default
+
+    strategy = get_import_strategy_from_string(normalized)
+    if strategy is None:
+        allowed = "/".join(allowed_import_strategy_values())
+        raise ValueError(f"strategy_override 必须为 {allowed}")
+    return strategy
+
+
+def validate_stored_knowledge_type(value: Any) -> KnowledgeType:
+    """校验写库 knowledge_type，仅允许合法落库类型。"""
+
+    if isinstance(value, KnowledgeType):
+        return value
+
+    resolved = get_knowledge_type_from_string(value)
+    if resolved is None:
+        allowed = "/".join(allowed_knowledge_type_values())
+        raise ValueError(f"knowledge_type 必须为 {allowed}")
+    return resolved
+
+
+def resolve_stored_knowledge_type(
+    value: Any,
+    *,
+    content: str = "",
+    allow_legacy: bool = False,
+    unknown_fallback: Optional[KnowledgeType] = None,
+) -> KnowledgeType:
+    """
+    将策略/字符串/旧值解析为合法落库类型。
+
+    `allow_legacy=True` 仅供迁移使用。
+    """
+
+    if isinstance(value, KnowledgeType):
+        return value
+
+    if isinstance(value, ImportStrategy):
+        if value == ImportStrategy.AUTO:
+            if not str(content or "").strip():
+                raise ValueError("knowledge_type=auto 需要 content 才能推断")
+            from .type_detection import detect_knowledge_type
+
+            return detect_knowledge_type(content)
+        return KnowledgeType(value.value)
+
+    raw = str(value or "").strip()
+    if not raw:
+        if str(content or "").strip():
+            from .type_detection import detect_knowledge_type
+
+            return detect_knowledge_type(content)
+        raise ValueError("knowledge_type 不能为空")
+
+    direct = get_knowledge_type_from_string(raw)
+    if direct is not None:
+        return direct
+
+    strategy = get_import_strategy_from_string(raw)
+    if strategy is not None:
+        return resolve_stored_knowledge_type(strategy, content=content)
+
+    if allow_legacy:
+        normalized = raw.lower()
+        if normalized == "imported":
+            return KnowledgeType.FACTUAL
+        if str(content or "").strip():
+            from .type_detection import detect_knowledge_type
+
+            detected = detect_knowledge_type(content)
+            if detected is not None:
+                return detected
+        if unknown_fallback is not None:
+            return unknown_fallback
+
+    allowed = "/".join(allowed_knowledge_type_values())
+    raise ValueError(f"非法 knowledge_type: {raw}（仅允许 {allowed}）")
+
+
+def should_extract_relations(knowledge_type: KnowledgeType) -> bool:
+    """判断是否应该做关系抽取。"""
+
+    return knowledge_type in [
+        KnowledgeType.STRUCTURED,
+        KnowledgeType.FACTUAL,
+        KnowledgeType.MIXED,
+    ]
+
+
+def get_default_chunk_size(knowledge_type: KnowledgeType) -> int:
+    """获取默认分块大小。"""
+
+    chunk_sizes = {
+        KnowledgeType.STRUCTURED: 300,
+        KnowledgeType.NARRATIVE: 800,
+        KnowledgeType.FACTUAL: 500,
+        KnowledgeType.QUOTE: 400,
+        KnowledgeType.MIXED: 500,
+    }
+    return chunk_sizes.get(knowledge_type, 500)
+
+
+def get_type_display_name(knowledge_type: KnowledgeType) -> str:
+    """获取知识类型中文名称。"""
+
+    display_names = {
+        KnowledgeType.STRUCTURED: "结构化知识",
+        KnowledgeType.NARRATIVE: "叙事性文本",
+        KnowledgeType.FACTUAL: "事实陈述",
+        KnowledgeType.QUOTE: "引用文本",
+        KnowledgeType.MIXED: "混合类型",
+    }
+    return display_names.get(knowledge_type, "未知类型")

plugins/A_memorix/core/storage/type_detection.py

@@ -0,0 +1,137 @@
+"""Heuristic detection for import strategies and stored knowledge types."""
+
+from __future__ import annotations
+
+import re
+from typing import Optional
+
+from .knowledge_types import (
+    ImportStrategy,
+    KnowledgeType,
+    parse_import_strategy,
+    resolve_stored_knowledge_type,
+)
+
+
+_NARRATIVE_MARKERS = [
+    r"然后",
+    r"接着",
+    r"于是",
+    r"后来",
+    r"最后",
+    r"突然",
+    r"一天",
+    r"曾经",
+    r"有一次",
+    r"从前",
+    r"说道",
+    r"问道",
+    r"想着",
+    r"觉得",
+]
+_FACTUAL_MARKERS = [
+    r"是",
+    r"有",
+    r"在",
+    r"为",
+    r"属于",
+    r"位于",
+    r"包含",
+    r"拥有",
+    r"成立于",
+    r"出生于",
+]
+
+
+def _non_empty_lines(content: str) -> list[str]:
+    return [line for line in str(content or "").splitlines() if line.strip()]
+
+
+def looks_like_structured_text(content: str) -> bool:
+    text = str(content or "").strip()
+    if "|" not in text or text.count("|") < 2:
+        return False
+    parts = text.split("|")
+    return len(parts) == 3 and all(part.strip() for part in parts)
+
+
+def looks_like_quote_text(content: str) -> bool:
+    lines = _non_empty_lines(content)
+    if len(lines) < 5:
+        return False
+    avg_len = sum(len(line) for line in lines) / len(lines)
+    return avg_len < 20
+
+
+def looks_like_narrative_text(content: str) -> bool:
+    text = str(content or "").strip()
+    if not text:
+        return False
+
+    narrative_score = sum(1 for marker in _NARRATIVE_MARKERS if re.search(marker, text))
+    has_dialogue = bool(re.search(r'["「『].*?["」』]', text))
+    has_chapter = any(token in text[:500] for token in ("Chapter", "CHAPTER", "###"))
+    return has_chapter or has_dialogue or narrative_score >= 2
+
+
+def looks_like_factual_text(content: str) -> bool:
+    text = str(content or "").strip()
+    if not text:
+        return False
+    if looks_like_structured_text(text) or looks_like_quote_text(text):
+        return False
+
+    factual_score = sum(1 for marker in _FACTUAL_MARKERS if re.search(r"\s*" + marker + r"\s*", text))
+    if factual_score <= 0:
+        return False
+
+    if len(text) <= 240:
+        return True
+    return factual_score >= 2 and not looks_like_narrative_text(text)
+
+
+def select_import_strategy(
+    content: str,
+    *,
+    override: Optional[str | ImportStrategy] = None,
+    chat_log: bool = False,
+) -> ImportStrategy:
+    """文本导入策略选择：override > quote > factual > narrative。"""
+
+    if chat_log:
+        return ImportStrategy.NARRATIVE
+
+    strategy = parse_import_strategy(override, default=ImportStrategy.AUTO)
+    if strategy != ImportStrategy.AUTO:
+        return strategy
+
+    if looks_like_quote_text(content):
+        return ImportStrategy.QUOTE
+    if looks_like_factual_text(content):
+        return ImportStrategy.FACTUAL
+    return ImportStrategy.NARRATIVE
+
+
+def detect_knowledge_type(content: str) -> KnowledgeType:
+    """自动检测落库 knowledge_type；无法可靠判断时回退 mixed。"""
+
+    text = str(content or "").strip()
+    if not text:
+        return KnowledgeType.MIXED
+    if looks_like_structured_text(text):
+        return KnowledgeType.STRUCTURED
+    if looks_like_quote_text(text):
+        return KnowledgeType.QUOTE
+    if looks_like_factual_text(text):
+        return KnowledgeType.FACTUAL
+    if looks_like_narrative_text(text):
+        return KnowledgeType.NARRATIVE
+    return KnowledgeType.MIXED
+
+
+def get_type_from_user_input(type_hint: Optional[str], content: str) -> KnowledgeType:
+    """优先使用显式 type_hint，否则自动检测。"""
+
+    if type_hint:
+        return resolve_stored_knowledge_type(type_hint, content=content)
+    return detect_knowledge_type(content)

plugins/A_memorix/core/storage/vector_store.py

@@ -0,0 +1,776 @@
+"""
+向量存储模块
+
+基于Faiss的高效向量存储与检索，支持SQ8量化、Append-Only磁盘存储和内存映射。
+"""
+
+import os
+import pickle
+import hashlib
+import shutil
+import time
+from pathlib import Path
+from typing import Optional, Union, Tuple, List, Dict, Set, Any
+import random
+import threading  # Added threading import
+
+import numpy as np
+
+try:
+    import faiss
+    HAS_FAISS = True
+except ImportError:
+    HAS_FAISS = False
+
+from src.common.logger import get_logger
+from ..utils.quantization import QuantizationType
+from ..utils.io import atomic_write, atomic_save_path
+
+logger = get_logger("A_Memorix.VectorStore")
+
+
+class VectorStore:
+    """
+    向量存储类 (SQ8 + Append-Only Disk)
+
+    特性：
+    - 索引: IndexIDMap2(IndexScalarQuantizer(QT_8bit))
+    - 存储: float16 on-disk binary (vectors.bin)
+    - 内存: 仅索引常驻 RAM (<512MB for 100k vectors)
+    - ID: SHA1-based stable int64 IDs
+    - 一致性: 强制 L2 Normalization (IP == Cosine)
+    """
+
+    # 默认训练触发阈值 (40 样本，过大可能导致小数据集不生效，过小可能量化退化)
+    DEFAULT_MIN_TRAIN = 40
+    # 强制训练样本量
+    TRAIN_SIZE = 10000
+    # 储水池采样上限 (流式处理前 50k 数据)
+    RESERVOIR_CAPACITY = 10000
+    RESERVOIR_SAMPLE_SCOPE = 50000
+
+    def __init__(
+        self,
+        dimension: int,
+        quantization_type: QuantizationType = QuantizationType.INT8,
+        index_type: str = "sq8",
+        data_dir: Optional[Union[str, Path]] = None,
+        use_mmap: bool = True,
+        buffer_size: int = 1024,
+    ):
+        if not HAS_FAISS:
+            raise ImportError("Faiss 未安装，请安装: pip install faiss-cpu")
+
+        self.dimension = dimension
+        self.data_dir = Path(data_dir) if data_dir else None
+        if self.data_dir:
+            self.data_dir.mkdir(parents=True, exist_ok=True)
+        if quantization_type != QuantizationType.INT8:
+            raise ValueError(
+                "vNext 仅支持 quantization_type=int8(SQ8)。"
+                " 请更新配置并执行 scripts/release_vnext_migrate.py migrate。"
+            )
+        normalized_index_type = str(index_type or "sq8").strip().lower()
+        if normalized_index_type not in {"sq8", "int8"}:
+            raise ValueError(
+                "vNext 仅支持 index_type=sq8。"
+                " 请更新配置并执行 scripts/release_vnext_migrate.py migrate。"
+            )
+        self.quantization_type = QuantizationType.INT8 
+        self.index_type = "sq8" 
+        self.buffer_size = buffer_size
+
+        self._index: Optional[faiss.IndexIDMap2] = None
+        self._init_index()
+
+        self._is_trained = False
+        self._vector_norm = "l2"
+        
+        # Fallback Index (Flat) - 用于在 SQ8 训练完成前提供检索能力
+        # 必须使用 IndexIDMap2 以保证 ID 与主索引一致
+        self._fallback_index: Optional[faiss.IndexIDMap2] = None
+        self._init_fallback_index()
+        
+        self._known_hashes: Set[str] = set()
+        self._deleted_ids: Set[int] = set()
+        
+        self._reservoir_buffer: List[np.ndarray] = []
+        self._seen_count_for_reservoir = 0
+
+        self._write_buffer_vecs: List[np.ndarray] = []
+        self._write_buffer_ids: List[int] = []
+
+        self._total_added = 0
+        self._total_deleted = 0
+        self._bin_count = 0 
+        
+        # Thread safety lock
+        self._lock = threading.RLock()
+
+        logger.info(f"VectorStore Init: dim={dimension}, SQ8 Mode, Append-Only Storage")
+
+    def _init_index(self):
+        """初始化空的 Faiss 索引"""
+        quantizer = faiss.IndexScalarQuantizer(
+            self.dimension, 
+            faiss.ScalarQuantizer.QT_8bit, 
+            faiss.METRIC_INNER_PRODUCT
+        )
+        self._index = faiss.IndexIDMap2(quantizer)
+        self._is_trained = False
+
+    def _init_fallback_index(self):
+        """初始化 Flat 回退索引"""
+        flat_index = faiss.IndexFlatIP(self.dimension)
+        self._fallback_index = faiss.IndexIDMap2(flat_index)
+        logger.debug("Fallback index (Flat) initialized.")
+
+    @staticmethod
+    def _generate_id(key: str) -> int:
+        """生成稳定的 int64 ID (SHA1 截断)"""
+        h = hashlib.sha1(key.encode("utf-8")).digest()
+        val = int.from_bytes(h[:8], byteorder="big", signed=False)
+        return val & 0x7FFFFFFFFFFFFFFF
+
+    @property
+    def _bin_path(self) -> Path:
+        return self.data_dir / "vectors.bin"
+    
+    @property
+    def _ids_bin_path(self) -> Path:
+        return self.data_dir / "vectors_ids.bin"
+
+    @property
+    def _int_to_str_map(self) -> Dict[int, str]:
+        """Lazy build volatile map from known hashes"""
+        # Note: This is read-heavy and cached, might need lock if _known_hashes updates concurrently
+        # But add/delete are now locked, so checking len mismatch is somewhat safe-ish for quick dirty cache
+        if not hasattr(self, "_cached_map") or len(self._cached_map) != len(self._known_hashes):
+            with self._lock: # Protect cache rebuild
+                 self._cached_map = {self._generate_id(k): k for k in self._known_hashes}
+        return self._cached_map
+
+    def add(self, vectors: np.ndarray, ids: List[str]) -> int:
+        with self._lock:
+            if vectors.shape[1] != self.dimension:
+                raise ValueError(f"Dimension mismatch: {vectors.shape[1]} vs {self.dimension}")
+
+            vectors = np.ascontiguousarray(vectors, dtype=np.float32)
+            faiss.normalize_L2(vectors)
+
+            processed_vecs = []
+            processed_int_ids = []
+            
+            for i, str_id in enumerate(ids):
+                if str_id in self._known_hashes:
+                    continue
+                
+                int_id = self._generate_id(str_id)
+                self._known_hashes.add(str_id)
+                
+                processed_vecs.append(vectors[i])
+                processed_int_ids.append(int_id)
+
+            if not processed_vecs:
+                return 0
+
+            batch_vecs = np.array(processed_vecs, dtype=np.float32)
+            batch_ids = np.array(processed_int_ids, dtype=np.int64)
+
+            self._write_buffer_vecs.append(batch_vecs)
+            self._write_buffer_ids.extend(processed_int_ids)
+
+            if len(self._write_buffer_ids) >= self.buffer_size:
+                self._flush_write_buffer_unlocked()
+
+            if not self._is_trained:
+                # 双写到回退索引
+                self._fallback_index.add_with_ids(batch_vecs, batch_ids)
+                
+                self._update_reservoir(batch_vecs)
+                # 这里的 TRAIN_SIZE 取默认 10k，或者根据当前数据量动态判断
+                if len(self._reservoir_buffer) >= 10000:
+                    logger.info(f"训练样本达到上限，开始训练...")
+                    self._train_and_replay_unlocked()
+
+            self._total_added += len(batch_ids)
+            return len(batch_ids)
+    
+    def _flush_write_buffer(self):
+        with self._lock:
+            self._flush_write_buffer_unlocked()
+
+    def _flush_write_buffer_unlocked(self):
+        if not self._write_buffer_vecs:
+            return
+
+        batch_vecs = np.concatenate(self._write_buffer_vecs, axis=0)
+        batch_ids = np.array(self._write_buffer_ids, dtype=np.int64)
+
+        vecs_fp16 = batch_vecs.astype(np.float16)
+        
+        with open(self._bin_path, "ab") as f:
+            f.write(vecs_fp16.tobytes())
+        
+        ids_bytes = batch_ids.astype('>i8').tobytes()
+        with open(self._ids_bin_path, "ab") as f:
+            f.write(ids_bytes)
+            
+        self._bin_count += len(batch_ids)
+
+        if self._is_trained and self._index.is_trained:
+            self._index.add_with_ids(batch_vecs, batch_ids)
+        else:
+            # 即使在 flush 时，如果未训练，也要同步到 fallback
+            self._fallback_index.add_with_ids(batch_vecs, batch_ids)
+
+        self._write_buffer_vecs.clear()
+        self._write_buffer_ids.clear()
+
+    def _update_reservoir(self, vectors: np.ndarray):
+        for vec in vectors:
+            self._seen_count_for_reservoir += 1
+            if len(self._reservoir_buffer) < self.RESERVOIR_CAPACITY:
+                self._reservoir_buffer.append(vec)
+            else:
+                if self._seen_count_for_reservoir <= self.RESERVOIR_SAMPLE_SCOPE:
+                    r = random.randint(0, self._seen_count_for_reservoir - 1)
+                    if r < self.RESERVOIR_CAPACITY:
+                        self._reservoir_buffer[r] = vec
+
+    def _train_and_replay(self):
+        with self._lock:
+            self._train_and_replay_unlocked()
+
+    def _train_and_replay_unlocked(self):
+        if not self._reservoir_buffer:
+            logger.warning("No training data available.")
+            return
+
+        train_data = np.array(self._reservoir_buffer, dtype=np.float32)
+        logger.info(f"Training Index with {len(train_data)} samples...")
+        
+        try:
+            self._index.train(train_data)
+        except Exception as e:
+            logger.error(f"SQ8 Training failed: {e}. Staying in fallback mode.")
+            return
+
+        self._is_trained = True
+        self._reservoir_buffer = []
+
+        logger.info("Replaying data from disk to populate index...")
+        try:
+            replay_count = self._replay_vectors_to_index()
+            # 只有当 replay 成功且数据量一致时，才释放回退索引
+            if self._index.ntotal >= self._bin_count:
+                logger.info(f"Replay successful ({self._index.ntotal}/{self._bin_count}). Releasing fallback index.")
+                self._fallback_index.reset()
+            else:
+                logger.warning(f"Replay count mismatch: {self._index.ntotal} vs {self._bin_count}. Keeping fallback index.")
+        except Exception as e:
+            logger.error(f"Replay failed: {e}. Keeping fallback index as backup.")
+
+    def _replay_vectors_to_index(self) -> int:
+        """从 vectors.bin 读取并添加到 index"""
+        if not self._bin_path.exists() or not self._ids_bin_path.exists():
+            return 0
+
+        vec_item_size = self.dimension * 2
+        id_item_size = 8
+        chunk_size = 10000 
+        
+        with open(self._bin_path, "rb") as f_vec, open(self._ids_bin_path, "rb") as f_id:
+            while True:
+                vec_data = f_vec.read(chunk_size * vec_item_size)
+                id_data = f_id.read(chunk_size * id_item_size)
+                
+                if not vec_data:
+                    break
+                
+                batch_fp16 = np.frombuffer(vec_data, dtype=np.float16).reshape(-1, self.dimension)
+                batch_fp32 = batch_fp16.astype(np.float32)
+                faiss.normalize_L2(batch_fp32)
+                
+                batch_ids = np.frombuffer(id_data, dtype='>i8').astype(np.int64)
+                
+                valid_mask = [id_ not in self._deleted_ids for id_ in batch_ids]
+                if not all(valid_mask):
+                    batch_fp32 = batch_fp32[valid_mask]
+                    batch_ids = batch_ids[valid_mask]
+                
+                if len(batch_ids) > 0:
+                    self._index.add_with_ids(batch_fp32, batch_ids)
+
+    def search(
+        self,
+        query: np.ndarray,
+        k: int = 10,
+        filter_deleted: bool = True,
+    ) -> Tuple[List[str], List[float]]:
+        query_local = np.array(query, dtype=np.float32, order="C", copy=True)
+        if query_local.ndim == 1:
+            got_dim = int(query_local.shape[0])
+            query_local = query_local.reshape(1, -1)
+        elif query_local.ndim == 2:
+            if query_local.shape[0] != 1:
+                raise ValueError(
+                    f"query embedding must have shape (D,) or (1, D), got {tuple(query_local.shape)}"
+                )
+            got_dim = int(query_local.shape[1])
+        else:
+            raise ValueError(
+                f"query embedding must have shape (D,) or (1, D), got {tuple(query_local.shape)}"
+            )
+
+        if got_dim != self.dimension:
+            raise ValueError(
+                f"query embedding dimension mismatch: expected={self.dimension} got={got_dim}"
+            )
+        if not np.all(np.isfinite(query_local)):
+            raise ValueError("query embedding contains non-finite values")
+
+        faiss.normalize_L2(query_local)
+
+        # 查询路径仅负责检索，不在此触发训练/回放。
+        # 训练/回放前置到 warmup_index()，并由插件启动阶段触发。
+        # Faiss 索引在并发 search 下可能出现阻塞，这里串行化检索调用保证稳定性。
+        with self._lock:
+            self._flush_write_buffer_unlocked()
+            search_index = self._index if (self._is_trained and self._index.ntotal > 0) else self._fallback_index
+            if search_index.ntotal == 0:
+                logger.warning("Indices are empty. No data to search.")
+                return [], []
+            # 执行检索
+            dists, ids = search_index.search(query_local, k * 2)
+        
+        # Faiss search 返回的是 (1, K) 的数组，取第一行
+        dists = dists[0]
+        ids = ids[0]
+        
+        results = []
+        for id_val, score in zip(ids, dists):
+            if id_val == -1: continue
+            if filter_deleted and id_val in self._deleted_ids:
+                continue
+            
+            str_id = self._int_to_str_map.get(id_val)
+            if str_id:
+                results.append((str_id, float(score)))
+
+        # Sort and trim just in case filtering reduced count
+        results.sort(key=lambda x: x[1], reverse=True)
+        results = results[:k]
+        
+        if not results:
+            return [], []
+            
+        return [r[0] for r in results], [r[1] for r in results]
+
+    def warmup_index(self, force_train: bool = True) -> Dict[str, Any]:
+        """
+        预热向量索引（训练/回放前置），避免首个线上查询触发重初始化。
+
+        Args:
+            force_train: 是否在满足阈值时强制训练 SQ8 索引
+
+        Returns:
+            预热状态摘要
+        """
+        started = time.perf_counter()
+        logger.info(f"metric.vector_index_prewarm_started=1 force_train={bool(force_train)}")
+
+        try:
+            with self._lock:
+                self._flush_write_buffer()
+
+                if self._bin_path.exists():
+                    self._bin_count = self._bin_path.stat().st_size // (self.dimension * 2)
+                else:
+                    self._bin_count = 0
+
+                needs_fallback_bootstrap = (
+                    self._bin_count > 0
+                    and self._fallback_index.ntotal == 0
+                    and (not self._is_trained or self._index.ntotal == 0)
+                )
+                if needs_fallback_bootstrap:
+                    self._bootstrap_fallback_from_disk()
+
+                min_train = max(1, int(getattr(self, "min_train_threshold", self.DEFAULT_MIN_TRAIN)))
+                needs_train = (
+                    bool(force_train)
+                    and self._bin_count >= min_train
+                    and not self._is_trained
+                )
+                if needs_train:
+                    self._force_train_small_data()
+
+                duration_ms = (time.perf_counter() - started) * 1000.0
+                summary = {
+                    "ok": True,
+                    "trained": bool(self._is_trained),
+                    "index_ntotal": int(self._index.ntotal),
+                    "fallback_ntotal": int(self._fallback_index.ntotal),
+                    "bin_count": int(self._bin_count),
+                    "duration_ms": duration_ms,
+                    "error": None,
+                }
+        except Exception as e:
+            duration_ms = (time.perf_counter() - started) * 1000.0
+            summary = {
+                "ok": False,
+                "trained": bool(self._is_trained),
+                "index_ntotal": int(self._index.ntotal) if self._index is not None else 0,
+                "fallback_ntotal": int(self._fallback_index.ntotal) if self._fallback_index is not None else 0,
+                "bin_count": int(getattr(self, "_bin_count", 0)),
+                "duration_ms": duration_ms,
+                "error": str(e),
+            }
+            logger.error(
+                "metric.vector_index_prewarm_fail=1 "
+                f"metric.vector_index_prewarm_duration_ms={duration_ms:.2f} "
+                f"error={e}"
+            )
+            return summary
+
+        logger.info(
+            "metric.vector_index_prewarm_success=1 "
+            f"metric.vector_index_prewarm_duration_ms={summary['duration_ms']:.2f} "
+            f"trained={summary['trained']} "
+            f"index_ntotal={summary['index_ntotal']} "
+            f"fallback_ntotal={summary['fallback_ntotal']} "
+            f"bin_count={summary['bin_count']}"
+        )
+        return summary
+
+    def _bootstrap_fallback_from_disk(self):
+        with self._lock:
+            self._bootstrap_fallback_from_disk_unlocked()
+
+    def _bootstrap_fallback_from_disk_unlocked(self):
+        """重启后自举：从磁盘 vectors.bin 加载数据到 fallback 索引"""
+        if not self._bin_path.exists() or not self._ids_bin_path.exists():
+            return
+
+        logger.info("Replaying all disk vectors to fallback index...")
+        vec_item_size = self.dimension * 2
+        id_item_size = 8
+        chunk_size = 10000 
+        
+        with open(self._bin_path, "rb") as f_vec, open(self._ids_bin_path, "rb") as f_id:
+            while True:
+                vec_data = f_vec.read(chunk_size * vec_item_size)
+                id_data = f_id.read(chunk_size * id_item_size)
+                if not vec_data: break
+                
+                batch_fp16 = np.frombuffer(vec_data, dtype=np.float16).reshape(-1, self.dimension)
+                batch_fp32 = batch_fp16.astype(np.float32)
+                faiss.normalize_L2(batch_fp32)
+                batch_ids = np.frombuffer(id_data, dtype='>i8').astype(np.int64)
+                
+                valid_mask = [id_ not in self._deleted_ids for id_ in batch_ids]
+                if any(valid_mask):
+                    self._fallback_index.add_with_ids(batch_fp32[valid_mask], batch_ids[valid_mask])
+        
+        logger.info(f"Fallback index self-bootstrapped with {self._fallback_index.ntotal} items.")
+
+    def _force_train_small_data(self):
+        with self._lock:
+            self._force_train_small_data_unlocked()
+
+    def _force_train_small_data_unlocked(self):
+        logger.info("Forcing training on small dataset...")
+        self._reservoir_buffer = [] 
+        
+        chunk_size = 10000
+        vec_item_size = self.dimension * 2
+        
+        with open(self._bin_path, "rb") as f:
+            while len(self._reservoir_buffer) < self.TRAIN_SIZE:
+                data = f.read(chunk_size * vec_item_size)
+                if not data: break
+                fp16 = np.frombuffer(data, dtype=np.float16).reshape(-1, self.dimension)
+                fp32 = fp16.astype(np.float32)
+                faiss.normalize_L2(fp32)
+                
+                for vec in fp32:
+                    self._reservoir_buffer.append(vec)
+                    if len(self._reservoir_buffer) >= self.TRAIN_SIZE:
+                        break
+        
+        self._train_and_replay_unlocked()
+
+    def delete(self, ids: List[str]) -> int:
+        with self._lock:
+            count = 0
+            for str_id in ids:
+                if str_id not in self._known_hashes:
+                    continue
+                int_id = self._generate_id(str_id)
+                if int_id not in self._deleted_ids:
+                    self._deleted_ids.add(int_id)
+                    if self._index.is_trained:
+                         self._index.remove_ids(np.array([int_id], dtype=np.int64))
+                    # 同步从 fallback 移除
+                    if self._fallback_index.ntotal > 0:
+                         self._fallback_index.remove_ids(np.array([int_id], dtype=np.int64))
+                    count += 1
+            self._total_deleted += count
+            
+            # Check GC
+            self._check_rebuild_needed()
+            return count
+
+    def _check_rebuild_needed(self):
+        """GC Excution Check"""
+        if self._bin_count == 0: return
+        ratio = len(self._deleted_ids) / self._bin_count
+        if ratio > 0.3 and len(self._deleted_ids) > 1000:
+            logger.info(f"Triggering GC/Rebuild (deleted ratio: {ratio:.2f})")
+            self.rebuild_index()
+
+    def rebuild_index(self):
+        """GC: 重建索引，压缩 bin 文件"""
+        with self._lock:
+            self._rebuild_index_locked()
+
+    def _rebuild_index_locked(self):
+        """实际 GC 重建逻辑。"""
+        logger.info("Starting Compaction (GC)...")
+        
+        tmp_bin = self.data_dir / "vectors.bin.tmp"
+        tmp_ids = self.data_dir / "vectors_ids.bin.tmp"
+        
+        vec_item_size = self.dimension * 2
+        id_item_size = 8
+        chunk_size = 10000
+        
+        new_count = 0
+        
+        # 1. Compact Files
+        with open(self._bin_path, "rb") as f_vec, open(self._ids_bin_path, "rb") as f_id, \
+             open(tmp_bin, "wb") as w_vec, open(tmp_ids, "wb") as w_id:
+            while True:
+                vec_data = f_vec.read(chunk_size * vec_item_size)
+                id_data = f_id.read(chunk_size * id_item_size)
+                if not vec_data: break
+                
+                batch_fp16 = np.frombuffer(vec_data, dtype=np.float16).reshape(-1, self.dimension)
+                batch_ids = np.frombuffer(id_data, dtype='>i8').astype(np.int64)
+                
+                keep_mask = [id_ not in self._deleted_ids for id_ in batch_ids]
+                
+                if any(keep_mask):
+                    keep_vecs = batch_fp16[keep_mask]
+                    keep_ids = batch_ids[keep_mask]
+                    
+                    w_vec.write(keep_vecs.tobytes())
+                    w_id.write(keep_ids.astype('>i8').tobytes())
+                    new_count += len(keep_ids)
+
+        # 2. Reset State & Atomic Swap
+        self._bin_count = new_count
+        
+        # Close current index
+        self._index.reset()
+        if self._fallback_index: self._fallback_index.reset() # Also clear fallback
+        self._is_trained = False
+        
+        # Swap files
+        shutil.move(str(tmp_bin), str(self._bin_path))
+        shutil.move(str(tmp_ids), str(self._ids_bin_path))
+        
+        # Reset Tombstones (Critical)
+        self._deleted_ids.clear()
+        
+        # 3. Reload/Rebuild Index (Fresh Train)
+        # We need to re-train because data distribution might have changed significantly after deletion
+        self._init_index()
+        self._init_fallback_index() # Re-init fallback too
+        self._force_train_small_data() # This will train and replay from the NEW compact file
+        
+        logger.info("Compaction Complete.")
+
+    def save(self, data_dir: Optional[Union[str, Path]] = None) -> None:
+        with self._lock:
+            if not data_dir:
+                data_dir = self.data_dir
+            if not data_dir:
+                raise ValueError("No data_dir")
+
+            data_dir = Path(data_dir)
+            data_dir.mkdir(parents=True, exist_ok=True)
+
+            self._flush_write_buffer_unlocked()
+
+            if self._is_trained:
+                index_path = data_dir / "vectors.index"
+                with atomic_save_path(index_path) as tmp:
+                    faiss.write_index(self._index, tmp)
+
+            meta = {
+                "dimension": self.dimension,
+                "quantization_type": self.quantization_type.value,
+                "is_trained": self._is_trained,
+                "vector_norm": self._vector_norm,
+                "deleted_ids": list(self._deleted_ids),
+                "known_hashes": list(self._known_hashes),
+            }
+            
+            with atomic_write(data_dir / "vectors_metadata.pkl", "wb") as f:
+                pickle.dump(meta, f)
+                
+            logger.info("VectorStore saved.")
+
+    def migrate_legacy_npy(self, data_dir: Optional[Union[str, Path]] = None) -> Dict[str, Any]:
+        """
+        离线迁移入口：将 legacy vectors.npy 转为 vNext 二进制格式。
+        """
+        with self._lock:
+            target_dir = Path(data_dir) if data_dir else self.data_dir
+            if target_dir is None:
+                raise ValueError("No data_dir")
+            target_dir = Path(target_dir)
+            npy_path = target_dir / "vectors.npy"
+            idx_path = target_dir / "vectors.index"
+            bin_path = target_dir / "vectors.bin"
+            ids_bin_path = target_dir / "vectors_ids.bin"
+            meta_path = target_dir / "vectors_metadata.pkl"
+
+            if not npy_path.exists():
+                return {"migrated": False, "reason": "npy_missing"}
+            if not meta_path.exists():
+                raise RuntimeError("legacy vectors.npy migration requires vectors_metadata.pkl")
+            if bin_path.exists() and ids_bin_path.exists():
+                return {"migrated": False, "reason": "bin_exists"}
+
+            # Reset in-memory state to avoid appending to stale runtime buffers.
+            self._known_hashes.clear()
+            self._deleted_ids.clear()
+            self._write_buffer_vecs.clear()
+            self._write_buffer_ids.clear()
+            self._init_index()
+            self._init_fallback_index()
+            self._is_trained = False
+            self._bin_count = 0
+
+            self._migrate_from_npy_unlocked(npy_path, idx_path, target_dir)
+            self.save(target_dir)
+            return {"migrated": True, "reason": "ok"}
+
+    def load(self, data_dir: Optional[Union[str, Path]] = None) -> None:
+        with self._lock:
+            if not data_dir: data_dir = self.data_dir
+            data_dir = Path(data_dir)
+            
+            npy_path = data_dir / "vectors.npy"
+            idx_path = data_dir / "vectors.index"
+            bin_path = data_dir / "vectors.bin"
+            
+            if npy_path.exists() and not bin_path.exists():
+                raise RuntimeError(
+                    "检测到 legacy vectors.npy，vNext 不再支持运行时自动迁移。"
+                    " 请先执行 scripts/release_vnext_migrate.py migrate。"
+                )
+
+            meta_path = data_dir / "vectors_metadata.pkl"
+            if not meta_path.exists():
+                logger.warning("No metadata found, initialized empty.")
+                return
+                
+            with open(meta_path, "rb") as f:
+                meta = pickle.load(f)
+                
+            if meta.get("vector_norm") != "l2":
+                logger.warning("Index IDMap2 version mismatch (L2 Norm), forcing rebuild...")
+                self._known_hashes = set(meta.get("ids", [])) | set(meta.get("known_hashes", []))
+                self._deleted_ids = set(meta.get("deleted_ids", []))
+                self._init_index()
+                self._force_train_small_data()
+                return
+
+            self._is_trained = meta.get("is_trained", False)
+            self._vector_norm = meta.get("vector_norm", "l2")
+            self._deleted_ids = set(meta.get("deleted_ids", []))
+            self._known_hashes = set(meta.get("known_hashes", []))
+            
+            if self._is_trained:
+                if idx_path.exists():
+                    try:
+                        self._index = faiss.read_index(str(idx_path))
+                        if not isinstance(self._index, faiss.IndexIDMap2):
+                            logger.warning("Loaded index type mismatch. Rebuilding...")
+                            self._init_index()
+                            self._force_train_small_data()
+                    except Exception as e:
+                         logger.error(f"Failed to load index: {e}. Rebuilding...")
+                         self._init_index()
+                         self._force_train_small_data()
+                else:
+                    logger.warning("Index file missing despite metadata indicating trained. Rebuilding from bin...")
+                    self._init_index()
+                    self._force_train_small_data()
+            
+            if bin_path.exists():
+                self._bin_count = bin_path.stat().st_size // (self.dimension * 2)
+
+    def _migrate_from_npy(self, npy_path, idx_path, data_dir):
+        with self._lock:
+            self._migrate_from_npy_unlocked(npy_path, idx_path, data_dir)
+
+    def _migrate_from_npy_unlocked(self, npy_path, idx_path, data_dir):
+        try:
+            arr = np.load(npy_path, mmap_mode="r")
+        except Exception:
+            arr = np.load(npy_path)
+            
+        meta_path = data_dir / "vectors_metadata.pkl"
+        old_ids = []
+        if meta_path.exists():
+            with open(meta_path, "rb") as f:
+                m = pickle.load(f)
+                old_ids = m.get("ids", [])
+                
+        if len(arr) != len(old_ids):
+            logger.error(f"Migration mismatch: arr {len(arr)} != ids {len(old_ids)}")
+            return
+
+        logger.info(f"Migrating {len(arr)} vectors...")
+        
+        chunk = 1000
+        for i in range(0, len(arr), chunk):
+            sub_arr = arr[i : i+chunk]
+            sub_ids = old_ids[i : i+chunk]
+            self.add(sub_arr, sub_ids)
+            
+        if not self._is_trained:
+            self._force_train_small_data()
+
+        shutil.move(str(npy_path), str(npy_path) + ".bak")
+        if idx_path.exists():
+            shutil.move(str(idx_path), str(idx_path) + ".bak")
+            
+        logger.info("Migration complete.")
+
+    def clear(self) -> None:
+        with self._lock:
+            self._ids_bin_path.unlink(missing_ok=True)
+            self._bin_path.unlink(missing_ok=True)
+            self._init_index()
+            self._known_hashes.clear()
+            self._deleted_ids.clear()
+            self._bin_count = 0
+            logger.info("VectorStore cleared.")
+
+    def has_data(self) -> bool:
+        return (self.data_dir / "vectors_metadata.pkl").exists()
+
+    @property
+    def num_vectors(self) -> int:
+        return len(self._known_hashes) - len(self._deleted_ids)
+
+    def __contains__(self, hash_value: str) -> bool:
+        """Check if a hash exists in the store"""
+        return hash_value in self._known_hashes and self._generate_id(hash_value) not in self._deleted_ids
+

plugins/A_memorix/core/strategies/base.py

@@ -0,0 +1,89 @@
+from abc import ABC, abstractmethod
+from typing import List, Dict, Any, Optional, Union
+from dataclasses import dataclass, field
+from enum import Enum
+import hashlib
+
+class KnowledgeType(str, Enum):
+    NARRATIVE = "narrative"
+    FACTUAL = "factual"
+    QUOTE = "quote"
+    MIXED = "mixed"
+
+@dataclass
+class SourceInfo:
+    file: str
+    offset_start: int
+    offset_end: int
+    checksum: str = ""
+
+@dataclass
+class ChunkContext:
+    chunk_id: str
+    index: int
+    context: Dict[str, Any] = field(default_factory=dict)
+    text: str = ""
+
+@dataclass
+class ChunkFlags:
+    verbatim: bool = False
+    requires_llm: bool = True
+
+@dataclass
+class ProcessedChunk:
+    type: KnowledgeType
+    source: SourceInfo
+    chunk: ChunkContext
+    data: Dict[str, Any] = field(default_factory=dict) # triples、events、verbatim_entities
+    flags: ChunkFlags = field(default_factory=ChunkFlags)
+
+    def to_dict(self) -> Dict:
+        return {
+            "type": self.type.value,
+            "source": {
+                "file": self.source.file,
+                "offset_start": self.source.offset_start,
+                "offset_end": self.source.offset_end,
+                "checksum": self.source.checksum
+            },
+            "chunk": {
+                "text": self.chunk.text,
+                "chunk_id": self.chunk.chunk_id,
+                "index": self.chunk.index,
+                "context": self.chunk.context
+            },
+            "data": self.data,
+            "flags": {
+                "verbatim": self.flags.verbatim,
+                "requires_llm": self.flags.requires_llm
+            }
+        }
+
+class BaseStrategy(ABC):
+    def __init__(self, filename: str):
+        self.filename = filename
+
+    @abstractmethod
+    def split(self, text: str) -> List[ProcessedChunk]:
+        """按策略将文本切分为块。"""
+        pass
+
+    @abstractmethod
+    async def extract(self, chunk: ProcessedChunk, llm_func=None) -> ProcessedChunk:
+        """从文本块中抽取结构化信息。"""
+        pass
+
+    def calculate_checksum(self, text: str) -> str:
+        return hashlib.sha256(text.encode("utf-8")).hexdigest()
+
+    def build_language_guard(self, text: str) -> str:
+        """
+        构建统一的输出语言约束。
+        不区分语言类型，仅要求抽取值保持原文语言，不做翻译。
+        """
+        _ = text  # 预留参数，便于后续按需扩展
+        return (
+            "Focus on the original source language. Keep extracted events, entities, predicates "
+            "and objects in the same language as the source text, preserve names/terms as-is, "
+            "and do not translate."
+        )

plugins/A_memorix/core/strategies/factual.py

@@ -0,0 +1,98 @@
+import re
+from typing import List, Dict, Any
+from .base import BaseStrategy, ProcessedChunk, KnowledgeType, SourceInfo, ChunkContext
+
+class FactualStrategy(BaseStrategy):
+    def split(self, text: str) -> List[ProcessedChunk]:
+        # 结构感知切分
+        lines = text.split('\n')
+        chunks = []
+        current_chunk_lines = []
+        current_len = 0
+        target_size = 600
+        
+        for i, line in enumerate(lines):
+            # 判断是否应当切分
+            # 若当前行为列表项/定义/表格行，则尽量不切分
+            is_structure = self._is_structural_line(line)
+            
+            current_len += len(line) + 1
+            current_chunk_lines.append(line)
+            
+            # 达到目标长度且不在紧凑结构块内时切分（过长时强制切分）
+            if current_len >= target_size and not is_structure:
+                 chunks.append(self._create_chunk(current_chunk_lines, len(chunks)))
+                 current_chunk_lines = []
+                 current_len = 0
+            elif current_len >= target_size * 2: # 超长时强制切分
+                 chunks.append(self._create_chunk(current_chunk_lines, len(chunks)))
+                 current_chunk_lines = []
+                 current_len = 0
+
+        if current_chunk_lines:
+            chunks.append(self._create_chunk(current_chunk_lines, len(chunks)))
+            
+        return chunks
+
+    def _is_structural_line(self, line: str) -> bool:
+        line = line.strip()
+        if not line: return False
+        # 列表项
+        if re.match(r'^[\-\*]\s+', line) or re.match(r'^\d+\.\s+', line):
+            return True
+        # 定义项（术语: 定义）
+        if re.match(r'^[^：:]+[：:].+', line):
+            return True
+        # 表格行（按 markdown 语法假设）
+        if line.startswith('|') and line.endswith('|'):
+            return True
+        return False
+
+    def _create_chunk(self, lines: List[str], index: int) -> ProcessedChunk:
+        text = "\n".join(lines)
+        return ProcessedChunk(
+            type=KnowledgeType.FACTUAL,
+            source=SourceInfo(
+                file=self.filename,
+                offset_start=0, # 简化处理：真实偏移跟踪需要额外状态
+                offset_end=0,
+                checksum=self.calculate_checksum(text)
+            ),
+            chunk=ChunkContext(
+                chunk_id=f"{self.filename}_{index}",
+                index=index,
+                text=text
+            )
+        )
+
+    async def extract(self, chunk: ProcessedChunk, llm_func=None) -> ProcessedChunk:
+        if not llm_func:
+            raise ValueError("LLM function required for Factual extraction")
+
+        language_guard = self.build_language_guard(chunk.chunk.text)
+        prompt = f"""You are a factual knowledge extraction engine.
+Extract factual triples and entities from the text.
+Preserve lists and definitions accurately.
+
+Language constraints:
+- {language_guard}
+- Preserve original names and domain terms exactly when possible.
+- JSON keys must stay exactly as: triples, entities, subject, predicate, object.
+
+Text:
+{chunk.chunk.text}
+
+Return ONLY valid JSON:
+{{
+  "triples": [
+    {{"subject": "Entity", "predicate": "Relationship", "object": "Entity"}}
+  ],
+  "entities": ["Entity1", "Entity2"]
+}}
+"""
+        result = await llm_func(prompt)
+        
+        # 结果保持原样存入 data，后续统一归一化流程会处理
+        # vector_store 侧期望关系字段为 subject/predicate/object 映射形式
+        chunk.data = result
+        return chunk

plugins/A_memorix/core/strategies/narrative.py

@@ -0,0 +1,126 @@
+import re
+from typing import List, Dict, Any
+from .base import BaseStrategy, ProcessedChunk, KnowledgeType, SourceInfo, ChunkContext
+
+class NarrativeStrategy(BaseStrategy):
+    def split(self, text: str) -> List[ProcessedChunk]:
+        scenes = self._split_into_scenes(text)
+        chunks = []
+        
+        for scene_idx, (scene_text, scene_title) in enumerate(scenes):
+             scene_chunks = self._sliding_window(scene_text, scene_title, scene_idx)
+             chunks.extend(scene_chunks)
+             
+        return chunks
+
+    def _split_into_scenes(self, text: str) -> List[tuple[str, str]]:
+        """按标题或分隔符把文本切分为场景。"""
+        # 简单启发式：按 markdown 标题或特定分隔符切分
+        # 该正则匹配以 #、Chapter 或 *** / === 开头的分隔行
+        # 该正则匹配以 #、Chapter 或 *** / === 开头的分隔行
+        scene_pattern_str = r'^(?:#{1,6}\s+.*|Chapter\s+\d+|^\*{3,}$|^={3,}$)'
+        
+        # 保留分隔符，以便识别场景起点
+        parts = re.split(f"({scene_pattern_str})", text, flags=re.MULTILINE)
+        
+        scenes = []
+        current_scene_title = "Start"
+        current_scene_content = []
+        
+        if parts and parts[0].strip() == "":
+            parts = parts[1:]
+            
+        for part in parts:
+            if re.match(scene_pattern_str, part, re.MULTILINE):
+                # 先保存上一段场景
+                if current_scene_content:
+                    scenes.append(("".join(current_scene_content), current_scene_title))
+                    current_scene_content = []
+                current_scene_title = part.strip()
+            else:
+                current_scene_content.append(part)
+                
+        if current_scene_content:
+             scenes.append(("".join(current_scene_content), current_scene_title))
+             
+        # 若未识别到场景，则把全文视作单一场景
+        if not scenes:
+            scenes = [(text, "Whole Text")]
+
+        return scenes
+
+    def _sliding_window(self, text: str, scene_id: str, scene_idx: int, window_size=800, overlap=200) -> List[ProcessedChunk]:
+        chunks = []
+        if len(text) <= window_size:
+            chunks.append(self._create_chunk(text, scene_id, scene_idx, 0, 0))
+            return chunks
+
+        stride = window_size - overlap
+        start = 0
+        local_idx = 0
+        while start < len(text):
+            end = min(start + window_size, len(text))
+            chunk_text = text[start:end]
+            
+            # 尽量对齐到最近换行，避免生硬截断句子
+            # 仅在未到文本尾部时进行回退
+            if end < len(text):
+                last_newline = chunk_text.rfind('\n')
+                if last_newline > window_size // 2: # 仅在回退距离可接受时启用
+                    end = start + last_newline + 1
+                    chunk_text = text[start:end]
+            
+            chunks.append(self._create_chunk(chunk_text, scene_id, scene_idx, local_idx, start))
+            
+            start += len(chunk_text) - overlap if end < len(text) else len(chunk_text)
+            local_idx += 1
+            
+        return chunks
+
+    def _create_chunk(self, text: str, scene_id: str, scene_idx: int, local_idx: int, offset: int) -> ProcessedChunk:
+        return ProcessedChunk(
+            type=KnowledgeType.NARRATIVE,
+            source=SourceInfo(
+                file=self.filename,
+                offset_start=offset,
+                offset_end=offset + len(text),
+                checksum=self.calculate_checksum(text)
+            ),
+            chunk=ChunkContext(
+                chunk_id=f"{self.filename}_{scene_idx}_{local_idx}",
+                index=local_idx,
+                text=text,
+                context={"scene_id": scene_id}
+            )
+        )
+
+    async def extract(self, chunk: ProcessedChunk, llm_func=None) -> ProcessedChunk:
+        if not llm_func:
+            raise ValueError("LLM function required for Narrative extraction")
+
+        language_guard = self.build_language_guard(chunk.chunk.text)
+        prompt = f"""You are a narrative knowledge extraction engine.
+Extract key events and character relations from the scene text.
+
+Language constraints:
+- {language_guard}
+- Preserve original names and terms exactly when possible.
+- JSON keys must stay exactly as: events, relations, subject, predicate, object.
+
+Scene:
+{chunk.chunk.context.get('scene_id')}
+
+Text:
+{chunk.chunk.text}
+
+Return ONLY valid JSON:
+{{
+  "events": ["event description 1", "event description 2"],
+  "relations": [
+    {{"subject": "CharacterA", "predicate": "relation", "object": "CharacterB"}}
+  ]
+}}
+"""
+        result = await llm_func(prompt)
+        chunk.data = result
+        return chunk

plugins/A_memorix/core/strategies/quote.py

@@ -0,0 +1,52 @@
+from typing import List, Dict, Any
+from .base import BaseStrategy, ProcessedChunk, KnowledgeType, SourceInfo, ChunkContext, ChunkFlags
+
+class QuoteStrategy(BaseStrategy):
+    def split(self, text: str) -> List[ProcessedChunk]:
+        # Split by double newlines (stanzas)
+        stanzas = text.split("\n\n")
+        chunks = []
+        offset = 0
+        
+        for idx, stanza in enumerate(stanzas):
+            if not stanza.strip():
+                offset += len(stanza) + 2
+                continue
+                
+            chunk = ProcessedChunk(
+                type=KnowledgeType.QUOTE,
+                source=SourceInfo(
+                    file=self.filename,
+                    offset_start=offset,
+                    offset_end=offset + len(stanza),
+                    checksum=self.calculate_checksum(stanza)
+                ),
+                chunk=ChunkContext(
+                    chunk_id=f"{self.filename}_{idx}",
+                    index=idx,
+                    text=stanza
+                ),
+                flags=ChunkFlags(
+                    verbatim=True,
+                    requires_llm=False # Default to no LLM, but can be overridden
+                )
+            )
+            chunks.append(chunk)
+            offset += len(stanza) + 2 # +2 for \n\n
+            
+        return chunks
+
+    async def extract(self, chunk: ProcessedChunk, llm_func=None) -> ProcessedChunk:
+        # For quotes, the text itself is the entity/knowledge
+        # We might use LLM to extract headers/metadata if requested, but core logic is pass-through
+        
+        # Treat the whole chunk text as a verbatim entity
+        chunk.data = {
+            "verbatim_entities": [chunk.chunk.text]
+        }
+        
+        if llm_func and chunk.flags.requires_llm:
+             # Optional: Extract metadata
+             pass
+             
+        return chunk

plugins/A_memorix/core/utils/__init__.py

@@ -0,0 +1,33 @@
+"""工具模块 - 哈希、监控等辅助功能"""
+
+from .hash import compute_hash, normalize_text
+from .monitor import MemoryMonitor
+from .quantization import quantize_vector, dequantize_vector
+from .time_parser import (
+    parse_query_datetime_to_timestamp,
+    parse_query_time_range,
+    parse_ingest_datetime_to_timestamp,
+    normalize_time_meta,
+    format_timestamp,
+)
+from .relation_write_service import RelationWriteService, RelationWriteResult
+from .relation_query import RelationQuerySpec, parse_relation_query_spec
+from .plugin_id_policy import PluginIdPolicy
+
+__all__ = [
+    "compute_hash",
+    "normalize_text",
+    "MemoryMonitor",
+    "quantize_vector",
+    "dequantize_vector",
+    "parse_query_datetime_to_timestamp",
+    "parse_query_time_range",
+    "parse_ingest_datetime_to_timestamp",
+    "normalize_time_meta",
+    "format_timestamp",
+    "RelationWriteService",
+    "RelationWriteResult",
+    "RelationQuerySpec",
+    "parse_relation_query_spec",
+    "PluginIdPolicy",
+]

plugins/A_memorix/core/utils/aggregate_query_service.py

@@ -0,0 +1,360 @@
+"""
+聚合查询服务：
+- 并发执行 search/time/episode 分支
+- 统一分支结果结构
+- 可选混合排序（Weighted RRF）
+"""
+
+from __future__ import annotations
+
+import asyncio
+from typing import Any, Awaitable, Callable, Dict, List, Optional, Tuple
+
+from src.common.logger import get_logger
+
+logger = get_logger("A_Memorix.AggregateQueryService")
+
+BranchRunner = Callable[[], Awaitable[Dict[str, Any]]]
+
+
+class AggregateQueryService:
+    """聚合查询执行服务（search/time/episode）。"""
+
+    def __init__(self, plugin_config: Optional[Any] = None):
+        self.plugin_config = plugin_config or {}
+
+    def _cfg(self, key: str, default: Any = None) -> Any:
+        getter = getattr(self.plugin_config, "get_config", None)
+        if callable(getter):
+            return getter(key, default)
+
+        current: Any = self.plugin_config
+        for part in key.split("."):
+            if isinstance(current, dict) and part in current:
+                current = current[part]
+            else:
+                return default
+        return current
+
+    @staticmethod
+    def _as_float(value: Any, default: float = 0.0) -> float:
+        try:
+            return float(value)
+        except Exception:
+            return float(default)
+
+    @staticmethod
+    def _as_int(value: Any, default: int = 0) -> int:
+        try:
+            return int(value)
+        except Exception:
+            return int(default)
+
+    def _rrf_k(self) -> float:
+        raw = self._cfg("retrieval.aggregate.rrf_k", 60.0)
+        value = self._as_float(raw, 60.0)
+        return max(1.0, value)
+
+    def _weights(self) -> Dict[str, float]:
+        defaults = {"search": 1.0, "time": 1.0, "episode": 1.0}
+        raw = self._cfg("retrieval.aggregate.weights", {})
+        if not isinstance(raw, dict):
+            return defaults
+
+        out = dict(defaults)
+        for key in ("search", "time", "episode"):
+            if key in raw:
+                out[key] = max(0.0, self._as_float(raw.get(key), defaults[key]))
+        return out
+
+    @staticmethod
+    def _normalize_branch_payload(
+        name: str,
+        payload: Optional[Dict[str, Any]],
+    ) -> Dict[str, Any]:
+        data = payload if isinstance(payload, dict) else {}
+        results_raw = data.get("results", [])
+        results = results_raw if isinstance(results_raw, list) else []
+        count = data.get("count")
+        if count is None:
+            count = len(results)
+        return {
+            "name": name,
+            "success": bool(data.get("success", False)),
+            "skipped": bool(data.get("skipped", False)),
+            "skip_reason": str(data.get("skip_reason", "") or "").strip(),
+            "error": str(data.get("error", "") or "").strip(),
+            "results": results,
+            "count": max(0, int(count)),
+            "elapsed_ms": max(0.0, float(data.get("elapsed_ms", 0.0) or 0.0)),
+            "content": str(data.get("content", "") or ""),
+            "query_type": str(data.get("query_type", "") or name),
+        }
+
+    @staticmethod
+    def _mix_key(item: Dict[str, Any], branch: str, rank: int) -> str:
+        item_type = str(item.get("type", "") or "").strip().lower()
+        if item_type == "episode":
+            episode_id = str(item.get("episode_id", "") or "").strip()
+            if episode_id:
+                return f"episode:{episode_id}"
+
+        item_hash = str(item.get("hash", "") or "").strip()
+        if item_hash:
+            return f"{item_type}:{item_hash}"
+
+        return f"{branch}:{item_type}:{rank}:{str(item.get('content', '') or '')[:80]}"
+
+    def _build_mixed_results(
+        self,
+        *,
+        branches: Dict[str, Dict[str, Any]],
+        top_k: int,
+    ) -> List[Dict[str, Any]]:
+        rrf_k = self._rrf_k()
+        weights = self._weights()
+        bucket: Dict[str, Dict[str, Any]] = {}
+
+        for branch_name, branch in branches.items():
+            if not branch.get("success", False):
+                continue
+            results = branch.get("results", [])
+            if not isinstance(results, list):
+                continue
+
+            weight = max(0.0, float(weights.get(branch_name, 1.0)))
+            for idx, item in enumerate(results, start=1):
+                if not isinstance(item, dict):
+                    continue
+                key = self._mix_key(item, branch_name, idx)
+                score = weight / (rrf_k + float(idx))
+                if key not in bucket:
+                    merged = dict(item)
+                    merged["fusion_score"] = 0.0
+                    merged["_source_branches"] = set()
+                    bucket[key] = merged
+
+                target = bucket[key]
+                target["fusion_score"] = float(target.get("fusion_score", 0.0)) + score
+                target["_source_branches"].add(branch_name)
+
+        mixed = list(bucket.values())
+        mixed.sort(
+            key=lambda x: (
+                -float(x.get("fusion_score", 0.0)),
+                str(x.get("type", "") or ""),
+                str(x.get("hash", "") or x.get("episode_id", "") or ""),
+            )
+        )
+
+        out: List[Dict[str, Any]] = []
+        for rank, item in enumerate(mixed[: max(1, int(top_k))], start=1):
+            merged = dict(item)
+            branches_set = merged.pop("_source_branches", set())
+            merged["source_branches"] = sorted(list(branches_set))
+            merged["rank"] = rank
+            out.append(merged)
+        return out
+
+    @staticmethod
+    def _status(branch: Dict[str, Any]) -> str:
+        if branch.get("skipped", False):
+            return "skipped"
+        if branch.get("success", False):
+            return "success"
+        return "failed"
+
+    def _build_summary(self, branches: Dict[str, Dict[str, Any]]) -> Dict[str, Dict[str, Any]]:
+        summary: Dict[str, Dict[str, Any]] = {}
+        for name, branch in branches.items():
+            status = self._status(branch)
+            summary[name] = {
+                "status": status,
+                "count": int(branch.get("count", 0) or 0),
+            }
+            if status == "skipped":
+                summary[name]["reason"] = str(branch.get("skip_reason", "") or "")
+            if status == "failed":
+                summary[name]["error"] = str(branch.get("error", "") or "")
+        return summary
+
+    def _build_content(
+        self,
+        *,
+        query: str,
+        branches: Dict[str, Dict[str, Any]],
+        errors: List[Dict[str, str]],
+        mixed_results: Optional[List[Dict[str, Any]]],
+    ) -> str:
+        lines: List[str] = [
+            f"🔀 聚合查询结果（query='{query or 'N/A'}'）",
+            "",
+            "分支状态：",
+        ]
+        for name in ("search", "time", "episode"):
+            branch = branches.get(name, {})
+            status = self._status(branch)
+            count = int(branch.get("count", 0) or 0)
+            line = f"- {name}: {status}, count={count}"
+            reason = str(branch.get("skip_reason", "") or "").strip()
+            err = str(branch.get("error", "") or "").strip()
+            if status == "skipped" and reason:
+                line += f" ({reason})"
+            if status == "failed" and err:
+                line += f" ({err})"
+            lines.append(line)
+
+        if errors:
+            lines.append("")
+            lines.append("错误：")
+            for item in errors[:6]:
+                lines.append(f"- {item.get('branch', 'unknown')}: {item.get('error', 'unknown error')}")
+
+        if mixed_results is not None:
+            lines.append("")
+            lines.append(f"🧩 混合结果（{len(mixed_results)} 条）：")
+            for idx, item in enumerate(mixed_results[:5], start=1):
+                src = ",".join(item.get("source_branches", []) or [])
+                if str(item.get("type", "") or "") == "episode":
+                    title = str(item.get("title", "") or "Untitled")
+                    lines.append(f"{idx}. 🧠 {title} [{src}]")
+                else:
+                    text = str(item.get("content", "") or "")
+                    if len(text) > 80:
+                        text = text[:80] + "..."
+                    lines.append(f"{idx}. {text} [{src}]")
+
+        return "\n".join(lines)
+
+    async def execute(
+        self,
+        *,
+        query: str,
+        top_k: int,
+        mix: bool,
+        mix_top_k: Optional[int],
+        time_from: Optional[str],
+        time_to: Optional[str],
+        search_runner: Optional[BranchRunner],
+        time_runner: Optional[BranchRunner],
+        episode_runner: Optional[BranchRunner],
+    ) -> Dict[str, Any]:
+        clean_query = str(query or "").strip()
+        safe_top_k = max(1, int(top_k))
+        safe_mix_top_k = max(1, int(mix_top_k if mix_top_k is not None else safe_top_k))
+
+        branches: Dict[str, Dict[str, Any]] = {}
+        errors: List[Dict[str, str]] = []
+        scheduled: List[Tuple[str, asyncio.Task]] = []
+
+        if clean_query:
+            if search_runner is not None:
+                scheduled.append(("search", asyncio.create_task(search_runner())))
+            else:
+                branches["search"] = self._normalize_branch_payload(
+                    "search",
+                    {"success": False, "error": "search runner unavailable", "results": []},
+                )
+        else:
+            branches["search"] = self._normalize_branch_payload(
+                "search",
+                {
+                    "success": False,
+                    "skipped": True,
+                    "skip_reason": "missing_query",
+                    "results": [],
+                    "count": 0,
+                },
+            )
+
+        if time_from or time_to:
+            if time_runner is not None:
+                scheduled.append(("time", asyncio.create_task(time_runner())))
+            else:
+                branches["time"] = self._normalize_branch_payload(
+                    "time",
+                    {"success": False, "error": "time runner unavailable", "results": []},
+                )
+        else:
+            branches["time"] = self._normalize_branch_payload(
+                "time",
+                {
+                    "success": False,
+                    "skipped": True,
+                    "skip_reason": "missing_time_window",
+                    "results": [],
+                    "count": 0,
+                },
+            )
+
+        if episode_runner is not None:
+            scheduled.append(("episode", asyncio.create_task(episode_runner())))
+        else:
+            branches["episode"] = self._normalize_branch_payload(
+                "episode",
+                {"success": False, "error": "episode runner unavailable", "results": []},
+            )
+
+        if scheduled:
+            done = await asyncio.gather(
+                *[task for _, task in scheduled],
+                return_exceptions=True,
+            )
+            for (branch_name, _), payload in zip(scheduled, done):
+                if isinstance(payload, Exception):
+                    logger.error("aggregate branch failed: branch=%s error=%s", branch_name, payload)
+                    normalized = self._normalize_branch_payload(
+                        branch_name,
+                        {
+                            "success": False,
+                            "error": str(payload),
+                            "results": [],
+                        },
+                    )
+                else:
+                    normalized = self._normalize_branch_payload(branch_name, payload)
+                branches[branch_name] = normalized
+
+        for name in ("search", "time", "episode"):
+            branch = branches.get(name)
+            if not branch:
+                continue
+            if branch.get("skipped", False):
+                continue
+            if not branch.get("success", False):
+                errors.append(
+                    {
+                        "branch": name,
+                        "error": str(branch.get("error", "") or "unknown error"),
+                    }
+                )
+
+        success = any(
+            bool(branches.get(name, {}).get("success", False))
+            for name in ("search", "time", "episode")
+        )
+        mixed_results: Optional[List[Dict[str, Any]]] = None
+        if mix:
+            mixed_results = self._build_mixed_results(branches=branches, top_k=safe_mix_top_k)
+
+        payload: Dict[str, Any] = {
+            "success": success,
+            "query_type": "aggregate",
+            "query": clean_query,
+            "top_k": safe_top_k,
+            "mix": bool(mix),
+            "mix_top_k": safe_mix_top_k,
+            "branches": branches,
+            "errors": errors,
+            "summary": self._build_summary(branches),
+        }
+        if mixed_results is not None:
+            payload["mixed_results"] = mixed_results
+
+        payload["content"] = self._build_content(
+            query=clean_query,
+            branches=branches,
+            errors=errors,
+            mixed_results=mixed_results,
+        )
+        return payload

plugins/A_memorix/core/utils/episode_retrieval_service.py

@@ -0,0 +1,182 @@
+"""Episode hybrid retrieval service."""
+
+from __future__ import annotations
+
+from typing import Any, Dict, List, Optional
+
+from src.common.logger import get_logger
+
+from ..retrieval import DualPathRetriever, TemporalQueryOptions
+
+logger = get_logger("A_Memorix.EpisodeRetrievalService")
+
+
+class EpisodeRetrievalService:
+    """Hybrid episode retrieval backed by lexical rows and evidence projection."""
+
+    _RRF_K = 60.0
+    _BRANCH_WEIGHTS = {
+        "lexical": 1.0,
+        "paragraph_evidence": 1.0,
+        "relation_evidence": 0.85,
+    }
+
+    def __init__(
+        self,
+        *,
+        metadata_store: Any,
+        retriever: Optional[DualPathRetriever] = None,
+    ) -> None:
+        self.metadata_store = metadata_store
+        self.retriever = retriever
+
+    async def query(
+        self,
+        *,
+        query: str = "",
+        top_k: int = 5,
+        time_from: Optional[float] = None,
+        time_to: Optional[float] = None,
+        person: Optional[str] = None,
+        source: Optional[str] = None,
+        include_paragraphs: bool = False,
+    ) -> List[Dict[str, Any]]:
+        clean_query = str(query or "").strip()
+        safe_top_k = max(1, int(top_k))
+        candidate_k = max(30, safe_top_k * 6)
+
+        branches: Dict[str, List[Dict[str, Any]]] = {
+            "lexical": self.metadata_store.query_episodes(
+                query=clean_query,
+                time_from=time_from,
+                time_to=time_to,
+                person=person,
+                source=source,
+                limit=(candidate_k if clean_query else safe_top_k),
+            )
+        }
+
+        if clean_query and self.retriever is not None:
+            try:
+                temporal = TemporalQueryOptions(
+                    time_from=time_from,
+                    time_to=time_to,
+                    person=person,
+                    source=source,
+                )
+                results = await self.retriever.retrieve(
+                    query=clean_query,
+                    top_k=candidate_k,
+                    temporal=temporal,
+                )
+            except Exception as exc:
+                logger.warning("episode evidence retrieval failed, fallback to lexical only: %s", exc)
+            else:
+                paragraph_rank_map: Dict[str, int] = {}
+                relation_rank_map: Dict[str, int] = {}
+                for rank, item in enumerate(results, start=1):
+                    hash_value = str(getattr(item, "hash_value", "") or "").strip()
+                    result_type = str(getattr(item, "result_type", "") or "").strip().lower()
+                    if not hash_value:
+                        continue
+                    if result_type == "paragraph" and hash_value not in paragraph_rank_map:
+                        paragraph_rank_map[hash_value] = rank
+                    elif result_type == "relation" and hash_value not in relation_rank_map:
+                        relation_rank_map[hash_value] = rank
+
+                if paragraph_rank_map:
+                    paragraph_rows = self.metadata_store.get_episode_rows_by_paragraph_hashes(
+                        list(paragraph_rank_map.keys()),
+                        source=source,
+                    )
+                    if paragraph_rows:
+                        branches["paragraph_evidence"] = self._rank_projected_rows(
+                            paragraph_rows,
+                            rank_map=paragraph_rank_map,
+                            support_key="matched_paragraph_hashes",
+                        )
+
+                if relation_rank_map:
+                    relation_rows = self.metadata_store.get_episode_rows_by_relation_hashes(
+                        list(relation_rank_map.keys()),
+                        source=source,
+                    )
+                    if relation_rows:
+                        branches["relation_evidence"] = self._rank_projected_rows(
+                            relation_rows,
+                            rank_map=relation_rank_map,
+                            support_key="matched_relation_hashes",
+                        )
+
+        fused = self._fuse_branches(branches, top_k=safe_top_k)
+        if include_paragraphs:
+            for item in fused:
+                item["paragraphs"] = self.metadata_store.get_episode_paragraphs(
+                    episode_id=str(item.get("episode_id") or ""),
+                    limit=50,
+                )
+        return fused
+
+    @staticmethod
+    def _rank_projected_rows(
+        rows: List[Dict[str, Any]],
+        *,
+        rank_map: Dict[str, int],
+        support_key: str,
+    ) -> List[Dict[str, Any]]:
+        sentinel = 10**9
+        ranked = [dict(item) for item in rows]
+
+        def _first_support_rank(item: Dict[str, Any]) -> int:
+            support_hashes = [str(x or "").strip() for x in (item.get(support_key) or [])]
+            ranks = [int(rank_map[h]) for h in support_hashes if h in rank_map]
+            return min(ranks) if ranks else sentinel
+
+        ranked.sort(
+            key=lambda item: (
+                _first_support_rank(item),
+                -int(item.get("matched_paragraph_count") or 0),
+                -float(item.get("updated_at") or 0.0),
+                str(item.get("episode_id") or ""),
+            )
+        )
+        return ranked
+
+    def _fuse_branches(
+        self,
+        branches: Dict[str, List[Dict[str, Any]]],
+        *,
+        top_k: int,
+    ) -> List[Dict[str, Any]]:
+        bucket: Dict[str, Dict[str, Any]] = {}
+        for branch_name, rows in branches.items():
+            weight = float(self._BRANCH_WEIGHTS.get(branch_name, 0.0) or 0.0)
+            if weight <= 0.0:
+                continue
+            for rank, row in enumerate(rows, start=1):
+                episode_id = str(row.get("episode_id", "") or "").strip()
+                if not episode_id:
+                    continue
+                if episode_id not in bucket:
+                    payload = dict(row)
+                    payload.pop("matched_paragraph_hashes", None)
+                    payload.pop("matched_relation_hashes", None)
+                    payload.pop("matched_paragraph_count", None)
+                    payload.pop("matched_relation_count", None)
+                    payload["_fusion_score"] = 0.0
+                    bucket[episode_id] = payload
+                bucket[episode_id]["_fusion_score"] = float(
+                    bucket[episode_id].get("_fusion_score", 0.0)
+                ) + weight / (self._RRF_K + float(rank))
+
+        out = list(bucket.values())
+        out.sort(
+            key=lambda item: (
+                -float(item.get("_fusion_score", 0.0)),
+                -float(item.get("updated_at") or 0.0),
+                str(item.get("episode_id") or ""),
+            )
+        )
+        for item in out:
+            item.pop("_fusion_score", None)
+        return out[: max(1, int(top_k))]

plugins/A_memorix/core/utils/hash.py

@@ -0,0 +1,129 @@
+"""
+哈希工具模块
+
+提供文本哈希计算功能，用于唯一标识和去重。
+"""
+
+import hashlib
+import re
+from typing import Union
+
+
+def compute_hash(text: str, hash_type: str = "sha256") -> str:
+    """
+    计算文本的哈希值
+
+    Args:
+        text: 输入文本
+        hash_type: 哈希算法类型（sha256, md5等）
+
+    Returns:
+        哈希值字符串
+    """
+    if hash_type == "sha256":
+        return hashlib.sha256(text.encode("utf-8")).hexdigest()
+    elif hash_type == "md5":
+        return hashlib.md5(text.encode("utf-8")).hexdigest()
+    else:
+        raise ValueError(f"不支持的哈希算法: {hash_type}")
+
+
+def normalize_text(text: str) -> str:
+    """
+    规范化文本用于哈希计算
+
+    执行以下操作：
+    - 去除首尾空白
+    - 统一换行符为\\n
+    - 压缩多个连续空格
+    - 去除不可见字符（保留换行和制表符）
+
+    Args:
+        text: 输入文本
+
+    Returns:
+        规范化后的文本
+    """
+    # 去除首尾空白
+    text = text.strip()
+
+    # 统一换行符
+    text = text.replace("\r\n", "\n").replace("\r", "\n")
+
+    # 压缩多个连续空格为一个（但保留换行和制表符）
+    text = re.sub(r"[^\S\n]+", " ", text)
+
+    return text
+
+
+def compute_paragraph_hash(paragraph: str) -> str:
+    """
+    计算段落的哈希值
+
+    Args:
+        paragraph: 段落文本
+
+    Returns:
+        段落哈希值（用于paragraph-前缀）
+    """
+    normalized = normalize_text(paragraph)
+    return compute_hash(normalized)
+
+
+def compute_entity_hash(entity: str) -> str:
+    """
+    计算实体的哈希值
+
+    Args:
+        entity: 实体名称
+
+    Returns:
+        实体哈希值（用于entity-前缀）
+    """
+    normalized = entity.strip().lower()
+    return compute_hash(normalized)
+
+
+def compute_relation_hash(relation: tuple) -> str:
+    """
+    计算关系的哈希值
+
+    Args:
+        relation: 关系元组 (subject, predicate, object)
+
+    Returns:
+        关系哈希值（用于relation-前缀）
+    """
+    # 将关系元组转为字符串
+    relation_str = str(tuple(relation))
+    return compute_hash(relation_str)
+
+
+def format_hash_key(hash_type: str, hash_value: str) -> str:
+    """
+    格式化哈希键
+
+    Args:
+        hash_type: 类型前缀（paragraph, entity, relation）
+        hash_value: 哈希值
+
+    Returns:
+        格式化的键（如 paragraph-abc123...）
+    """
+    return f"{hash_type}-{hash_value}"
+
+
+def parse_hash_key(key: str) -> tuple[str, str]:
+    """
+    解析哈希键
+
+    Args:
+        key: 格式化的键（如 paragraph-abc123...）
+
+    Returns:
+        (类型, 哈希值) 元组
+    """
+    parts = key.split("-", 1)
+    if len(parts) != 2:
+        raise ValueError(f"无效的哈希键格式: {key}")
+    return parts[0], parts[1]

plugins/A_memorix/core/utils/import_payloads.py

@@ -0,0 +1,110 @@
+"""Shared import payload normalization helpers."""
+
+from __future__ import annotations
+
+from typing import Any, Dict, List, Optional
+
+from ..storage import KnowledgeType, resolve_stored_knowledge_type
+from .time_parser import normalize_time_meta
+
+
+def _normalize_entities(raw_entities: Any) -> List[str]:
+    if not isinstance(raw_entities, list):
+        return []
+    out: List[str] = []
+    seen = set()
+    for item in raw_entities:
+        name = str(item or "").strip()
+        if not name:
+            continue
+        key = name.lower()
+        if key in seen:
+            continue
+        seen.add(key)
+        out.append(name)
+    return out
+
+
+def _normalize_relations(raw_relations: Any) -> List[Dict[str, str]]:
+    if not isinstance(raw_relations, list):
+        return []
+    out: List[Dict[str, str]] = []
+    for item in raw_relations:
+        if not isinstance(item, dict):
+            continue
+        subject = str(item.get("subject", "")).strip()
+        predicate = str(item.get("predicate", "")).strip()
+        obj = str(item.get("object", "")).strip()
+        if not (subject and predicate and obj):
+            continue
+        out.append(
+            {
+                "subject": subject,
+                "predicate": predicate,
+                "object": obj,
+            }
+        )
+    return out
+
+
+def normalize_paragraph_import_item(
+    item: Any,
+    *,
+    default_source: str,
+) -> Dict[str, Any]:
+    """Normalize one paragraph import item from text/json payloads."""
+
+    if isinstance(item, str):
+        content = str(item)
+        knowledge_type = resolve_stored_knowledge_type(None, content=content)
+        return {
+            "content": content,
+            "knowledge_type": knowledge_type.value,
+            "source": str(default_source or "").strip(),
+            "time_meta": None,
+            "entities": [],
+            "relations": [],
+        }
+
+    if not isinstance(item, dict) or "content" not in item:
+        raise ValueError("段落项必须为字符串或包含 content 的对象")
+
+    content = str(item.get("content", "") or "")
+    if not content.strip():
+        raise ValueError("段落 content 不能为空")
+
+    raw_time_meta = {
+        "event_time": item.get("event_time"),
+        "event_time_start": item.get("event_time_start"),
+        "event_time_end": item.get("event_time_end"),
+        "time_range": item.get("time_range"),
+        "time_granularity": item.get("time_granularity"),
+        "time_confidence": item.get("time_confidence"),
+    }
+    time_meta_field = item.get("time_meta")
+    if isinstance(time_meta_field, dict):
+        raw_time_meta.update(time_meta_field)
+
+    knowledge_type_raw = item.get("knowledge_type")
+    if knowledge_type_raw is None:
+        knowledge_type_raw = item.get("type")
+    knowledge_type = resolve_stored_knowledge_type(knowledge_type_raw, content=content)
+    source = str(item.get("source") or default_source or "").strip()
+    if not source:
+        source = str(default_source or "").strip()
+
+    normalized_time_meta = normalize_time_meta(raw_time_meta)
+    return {
+        "content": content,
+        "knowledge_type": knowledge_type.value,
+        "source": source,
+        "time_meta": normalized_time_meta if normalized_time_meta else None,
+        "entities": _normalize_entities(item.get("entities")),
+        "relations": _normalize_relations(item.get("relations")),
+    }
+
+
+def normalize_summary_knowledge_type(value: Any) -> KnowledgeType:
+    """Normalize config-driven summary knowledge type."""
+
+    return resolve_stored_knowledge_type(value, content="")

plugins/A_memorix/core/utils/io.py

@@ -0,0 +1,84 @@
+"""
+IO Utilities
+
+提供原子文件写入等IO辅助功能。
+"""
+
+import os
+import shutil
+import contextlib
+from pathlib import Path
+from typing import Union
+
+@contextlib.contextmanager
+def atomic_write(file_path: Union[str, Path], mode: str = "w", encoding: str = None, **kwargs):
+    """
+    原子文件写入上下文管理器
+    
+    原理：
+    1. 写入 .tmp 临时文件
+    2. 写入成功后，使用 os.replace 原子替换目标文件
+    3. 如果失败，自动删除临时文件
+    
+    Args:
+        file_path: 目标文件路径
+        mode: 打开模式 ('w', 'wb' 等)
+        encoding: 编码
+        **kwargs: 传给 open() 的其他参数
+    """
+    path = Path(file_path)
+    # 确保父目录存在
+    path.parent.mkdir(parents=True, exist_ok=True)
+    
+    # 临时文件路径
+    tmp_path = path.with_suffix(path.suffix + ".tmp")
+    
+    try:
+        with open(tmp_path, mode, encoding=encoding, **kwargs) as f:
+            yield f
+            
+            # 确保写入磁盘
+            f.flush()
+            os.fsync(f.fileno())
+            
+        # 原子替换 (Windows下可能需要先删除目标文件，但 os.replace 在 Py3.3+ 尽可能原子)
+        # 注意: Windows 上如果有其他进程占用文件，os.replace 可能会失败
+        os.replace(tmp_path, path)
+        
+    except Exception as e:
+        # 清理临时文件
+        if tmp_path.exists():
+            try:
+                os.remove(tmp_path)
+            except:
+                pass
+        raise e
+
+@contextlib.contextmanager
+def atomic_save_path(file_path: Union[str, Path]):
+    """
+    提供临时路径用于原子保存 (针对只接受路径的API，如Faiss)
+    
+    Args:
+        file_path: 最终目标路径
+        
+    Yields:
+        tmp_path: 临时文件路径 (str)
+    """
+    path = Path(file_path)
+    path.parent.mkdir(parents=True, exist_ok=True)
+    tmp_path = path.with_suffix(path.suffix + ".tmp")
+    
+    try:
+        yield str(tmp_path)
+        
+        if Path(tmp_path).exists():
+            os.replace(tmp_path, path)
+            
+    except Exception as e:
+        if Path(tmp_path).exists():
+            try:
+                os.remove(tmp_path)
+            except:
+                pass
+        raise e

plugins/A_memorix/core/utils/matcher.py

@@ -0,0 +1,89 @@
+"""
+高效文本匹配工具模块
+
+实现 Aho-Corasick 算法用于多模式匹配。
+"""
+
+from typing import List, Dict, Tuple, Set, Any
+from collections import deque
+
+
+class AhoCorasick:
+    """
+    Aho-Corasick 自动机实现高效多模式匹配
+    """
+
+    def __init__(self):
+        # next_states[state][char] = next_state
+        self.next_states: List[Dict[str, int]] = [{}]
+        # fail[state] = fail_state
+        self.fail: List[int] = [0]
+        # output[state] = set of patterns ending at this state
+        self.output: List[Set[str]] = [set()]
+        self.patterns: Set[str] = set()
+
+    def add_pattern(self, pattern: str):
+        """添加模式"""
+        if not pattern:
+            return
+        self.patterns.add(pattern)
+        state = 0
+        for char in pattern:
+            if char not in self.next_states[state]:
+                new_state = len(self.next_states)
+                self.next_states[state][char] = new_state
+                self.next_states.append({})
+                self.fail.append(0)
+                self.output.append(set())
+            state = self.next_states[state][char]
+        self.output[state].add(pattern)
+
+    def build(self):
+        """构建失败指针"""
+        queue = deque()
+        # 处理第一层
+        for char, state in self.next_states[0].items():
+            queue.append(state)
+            self.fail[state] = 0
+
+        while queue:
+            r = queue.popleft()
+            for char, s in self.next_states[r].items():
+                queue.append(s)
+                # 找到失败路径
+                state = self.fail[r]
+                while char not in self.next_states[state] and state != 0:
+                    state = self.fail[state]
+                self.fail[s] = self.next_states[state].get(char, 0)
+                # 合并输出
+                self.output[s].update(self.output[self.fail[s]])
+
+    def search(self, text: str) -> List[Tuple[int, str]]:
+        """
+        在文本中搜索所有模式
+        
+        Returns:
+            [(结束索引, 匹配到的模式), ...]
+        """
+        state = 0
+        results = []
+        for i, char in enumerate(text):
+            while char not in self.next_states[state] and state != 0:
+                state = self.fail[state]
+            state = self.next_states[state].get(char, 0)
+            for pattern in self.output[state]:
+                results.append((i, pattern))
+        return results
+
+    def find_all(self, text: str) -> Dict[str, int]:
+        """
+        查找并统计所有模式出现次数
+        
+        Returns:
+            {模式: 出现次数}
+        """
+        results = self.search(text)
+        stats = {}
+        for _, pattern in results:
+            stats[pattern] = stats.get(pattern, 0) + 1
+        return stats

plugins/A_memorix/core/utils/monitor.py

@@ -0,0 +1,189 @@
+"""
+内存监控模块
+
+提供内存使用监控和预警功能。
+"""
+
+import gc
+import threading
+import time
+from typing import Callable, Optional
+
+try:
+    import psutil
+    HAS_PSUTIL = True
+except ImportError:
+    HAS_PSUTIL = False
+
+from src.common.logger import get_logger
+
+logger = get_logger("A_Memorix.MemoryMonitor")
+
+
+class MemoryMonitor:
+    """
+    内存监控器
+
+    功能：
+    - 实时监控内存使用
+    - 超过阈值时触发警告
+    - 支持自动垃圾回收
+    """
+
+    def __init__(
+        self,
+        max_memory_mb: int,
+        warning_threshold: float = 0.9,
+        check_interval: float = 10.0,
+        enable_auto_gc: bool = True,
+    ):
+        """
+        初始化内存监控器
+
+        Args:
+            max_memory_mb: 最大内存限制（MB）
+            warning_threshold: 警告阈值（0-1之间，默认0.9表示90%）
+            check_interval: 检查间隔（秒）
+            enable_auto_gc: 是否启用自动垃圾回收
+        """
+        self.max_memory_mb = max_memory_mb
+        self.warning_threshold = warning_threshold
+        self.check_interval = check_interval
+        self.enable_auto_gc = enable_auto_gc
+
+        self._running = False
+        self._thread: Optional[threading.Thread] = None
+        self._callbacks: list[Callable[[float, float], None]] = []
+
+    def start(self):
+        """启动监控"""
+        if self._running:
+            logger.warning("内存监控已在运行")
+            return
+
+        if not HAS_PSUTIL:
+            logger.warning("psutil 未安装，内存监控功能不可用")
+            return
+
+        self._running = True
+        self._thread = threading.Thread(target=self._monitor_loop, daemon=True)
+        self._thread.start()
+        logger.info(f"内存监控已启动 (限制: {self.max_memory_mb}MB)")
+
+    def stop(self):
+        """停止监控"""
+        if not self._running:
+            return
+
+        self._running = False
+        if self._thread:
+            self._thread.join(timeout=5.0)
+        logger.info("内存监控已停止")
+
+    def register_callback(self, callback: Callable[[float, float], None]):
+        """
+        注册内存超限回调函数
+
+        Args:
+            callback: 回调函数，接收 (当前使用MB, 限制MB) 参数
+        """
+        self._callbacks.append(callback)
+
+    def get_current_memory_mb(self) -> float:
+        """
+        获取当前进程内存使用量（MB）
+
+        Returns:
+            内存使用量（MB）
+        """
+        if not HAS_PSUTIL:
+            # 降级方案：使用内置函数
+            import sys
+            return sys.getsizeof(gc.get_objects()) / 1024 / 1024
+
+        process = psutil.Process()
+        return process.memory_info().rss / 1024 / 1024
+
+    def get_memory_usage_ratio(self) -> float:
+        """
+        获取内存使用率
+
+        Returns:
+            使用率（0-1之间）
+        """
+        current = self.get_current_memory_mb()
+        return current / self.max_memory_mb if self.max_memory_mb > 0 else 0
+
+    def _monitor_loop(self):
+        """监控循环"""
+        while self._running:
+            try:
+                current_mb = self.get_current_memory_mb()
+                ratio = current_mb / self.max_memory_mb if self.max_memory_mb > 0 else 0
+
+                # 检查是否超过阈值
+                if ratio >= self.warning_threshold:
+                    logger.warning(
+                        f"内存使用率过高: {current_mb:.1f}MB / {self.max_memory_mb}MB "
+                        f"({ratio*100:.1f}%)"
+                    )
+
+                    # 触发回调
+                    for callback in self._callbacks:
+                        try:
+                            callback(current_mb, self.max_memory_mb)
+                        except Exception as e:
+                            logger.error(f"内存回调执行失败: {e}")
+
+                    # 自动垃圾回收
+                    if self.enable_auto_gc:
+                        before = self.get_current_memory_mb()
+                        gc.collect()
+                        after = self.get_current_memory_mb()
+                        freed = before - after
+                        if freed > 1:  # 释放超过1MB才记录
+                            logger.info(f"垃圾回收释放: {freed:.1f}MB")
+
+                # 定期垃圾回收（即使未超限）
+                elif self.enable_auto_gc and int(time.time()) % 60 == 0:
+                    gc.collect()
+
+            except Exception as e:
+                logger.error(f"内存监控出错: {e}")
+
+            # 等待下次检查
+            time.sleep(self.check_interval)
+
+    def __enter__(self):
+        """上下文管理器入口"""
+        self.start()
+        return self
+
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        """上下文管理器出口"""
+        self.stop()
+
+
+def get_memory_info() -> dict:
+    """
+    获取系统内存信息
+
+    Returns:
+        内存信息字典
+    """
+    if not HAS_PSUTIL:
+        return {"error": "psutil 未安装"}
+
+    try:
+        mem = psutil.virtual_memory()
+        process = psutil.Process()
+
+        return {
+            "system_total_gb": mem.total / 1024 / 1024 / 1024,
+            "system_available_gb": mem.available / 1024 / 1024 / 1024,
+            "system_usage_percent": mem.percent,
+            "process_mb": process.memory_info().rss / 1024 / 1024,
+            "process_percent": (process.memory_info().rss / mem.total) * 100,
+        }
+    except Exception as e:
+        return {"error": str(e)}

plugins/A_memorix/core/utils/path_fallback_service.py

@@ -0,0 +1,165 @@
+"""Shared path-fallback helpers for search post-processing."""
+
+from __future__ import annotations
+
+import hashlib
+from typing import Any, Dict, List, Optional, Sequence, Tuple
+
+from ..retrieval.dual_path import RetrievalResult
+
+
+def extract_entities(query: str, graph_store: Any) -> List[str]:
+    """Extract up to two graph nodes from a query using n-gram matching."""
+    if not graph_store:
+        return []
+
+    text = str(query or "").strip()
+    if not text:
+        return []
+
+    # Keep the heuristic aligned with previous legacy behavior.
+    tokens = (
+        text.replace("?", " ")
+        .replace("!", " ")
+        .replace(".", " ")
+        .split()
+    )
+    if not tokens:
+        return []
+
+    found_entities = set()
+    skip_indices = set()
+    max_n = min(4, len(tokens))
+
+    for size in range(max_n, 0, -1):
+        for i in range(len(tokens) - size + 1):
+            if any(idx in skip_indices for idx in range(i, i + size)):
+                continue
+            span = " ".join(tokens[i : i + size])
+            matched_node = graph_store.find_node(span, ignore_case=True)
+            if not matched_node:
+                continue
+            found_entities.add(matched_node)
+            for idx in range(i, i + size):
+                skip_indices.add(idx)
+
+    return list(found_entities)
+
+
+def find_paths_between_entities(
+    start_node: str,
+    end_node: str,
+    graph_store: Any,
+    metadata_store: Any,
+    *,
+    max_depth: int = 3,
+    max_paths: int = 5,
+) -> List[Dict[str, Any]]:
+    """Find and enrich indirect paths between two nodes."""
+    if not graph_store or not metadata_store:
+        return []
+
+    try:
+        paths = graph_store.find_paths(
+            start_node,
+            end_node,
+            max_depth=max_depth,
+            max_paths=max_paths,
+        )
+    except Exception:
+        return []
+
+    if not paths:
+        return []
+
+    edge_cache: Dict[Tuple[str, str], Tuple[str, str]] = {}
+    formatted_paths: List[Dict[str, Any]] = []
+
+    for path_nodes in paths:
+        if not isinstance(path_nodes, Sequence) or len(path_nodes) < 2:
+            continue
+
+        path_desc: List[str] = []
+        for i in range(len(path_nodes) - 1):
+            u = str(path_nodes[i])
+            v = str(path_nodes[i + 1])
+
+            cache_key = tuple(sorted((u, v)))
+            if cache_key in edge_cache:
+                pred, direction = edge_cache[cache_key]
+            else:
+                pred = "related"
+                direction = "->"
+                rels = metadata_store.get_relations(subject=u, object=v)
+                if not rels:
+                    rels = metadata_store.get_relations(subject=v, object=u)
+                    direction = "<-"
+                if rels:
+                    best_rel = max(rels, key=lambda x: x.get("confidence", 1.0))
+                    pred = str(best_rel.get("predicate", "related") or "related")
+                edge_cache[cache_key] = (pred, direction)
+
+            step_str = f"-[{pred}]->" if direction == "->" else f"<-[{pred}]-"
+            path_desc.append(step_str)
+
+        full_path_str = str(path_nodes[0])
+        for i, step in enumerate(path_desc):
+            full_path_str += f" {step} {path_nodes[i + 1]}"
+
+        formatted_paths.append(
+            {
+                "nodes": list(path_nodes),
+                "description": full_path_str,
+            }
+        )
+
+    return formatted_paths
+
+
+def find_paths_from_query(
+    query: str,
+    graph_store: Any,
+    metadata_store: Any,
+    *,
+    max_depth: int = 3,
+    max_paths: int = 5,
+) -> List[Dict[str, Any]]:
+    """Extract entities from query and resolve indirect paths."""
+    entities = extract_entities(query, graph_store)
+    if len(entities) != 2:
+        return []
+    return find_paths_between_entities(
+        entities[0],
+        entities[1],
+        graph_store,
+        metadata_store,
+        max_depth=max_depth,
+        max_paths=max_paths,
+    )
+
+
+def to_retrieval_results(paths: Sequence[Dict[str, Any]]) -> List[RetrievalResult]:
+    """Convert path results into retrieval results for the unified pipeline."""
+    converted: List[RetrievalResult] = []
+    for item in paths:
+        description = str(item.get("description", "")).strip()
+        if not description:
+            continue
+        hash_seed = description.encode("utf-8")
+        path_hash = f"path_{hashlib.sha1(hash_seed).hexdigest()}"
+        converted.append(
+            RetrievalResult(
+                hash_value=path_hash,
+                content=f"[Indirect Relation] {description}",
+                score=0.95,
+                result_type="relation",
+                source="graph_path",
+                metadata={
+                    "source": "graph_path",
+                    "is_indirect": True,
+                    "nodes": list(item.get("nodes", [])),
+                },
+            )
+        )
+    return converted
+

plugins/A_memorix/core/utils/person_profile_service.py

@@ -0,0 +1,495 @@
+"""
+人物画像服务
+
+主链路：
+person_id -> 用户名/别名 -> 图谱关系 + 向量证据 -> 证据总结画像 -> 快照版本化存储
+"""
+
+import json
+import time
+from typing import Any, Dict, List, Optional, Tuple
+
+from src.common.logger import get_logger
+from src.common.database.database_model import PersonInfo
+
+from ..embedding import EmbeddingAPIAdapter
+from ..retrieval import (
+    DualPathRetriever,
+    RetrievalStrategy,
+    DualPathRetrieverConfig,
+    SparseBM25Config,
+    FusionConfig,
+    GraphRelationRecallConfig,
+)
+from ..storage import MetadataStore, GraphStore, VectorStore
+
+logger = get_logger("A_Memorix.PersonProfileService")
+
+
+class PersonProfileService:
+    """人物画像聚合/刷新服务。"""
+
+    def __init__(
+        self,
+        metadata_store: MetadataStore,
+        graph_store: Optional[GraphStore] = None,
+        vector_store: Optional[VectorStore] = None,
+        embedding_manager: Optional[EmbeddingAPIAdapter] = None,
+        sparse_index: Any = None,
+        plugin_config: Optional[dict] = None,
+        retriever: Optional[DualPathRetriever] = None,
+    ):
+        self.metadata_store = metadata_store
+        self.graph_store = graph_store
+        self.vector_store = vector_store
+        self.embedding_manager = embedding_manager
+        self.sparse_index = sparse_index
+        self.plugin_config = plugin_config or {}
+        self.retriever = retriever or self._build_retriever()
+
+    def _cfg(self, key: str, default: Any = None) -> Any:
+        """读取嵌套配置。"""
+        if not isinstance(self.plugin_config, dict):
+            return default
+        current: Any = self.plugin_config
+        for part in key.split("."):
+            if isinstance(current, dict) and part in current:
+                current = current[part]
+            else:
+                return default
+        return current
+
+    def _build_retriever(self) -> Optional[DualPathRetriever]:
+        """按需构建检索器（无依赖时返回 None）。"""
+        if not all(
+            [
+                self.vector_store is not None,
+                self.graph_store is not None,
+                self.metadata_store is not None,
+                self.embedding_manager is not None,
+            ]
+        ):
+            return None
+        try:
+            sparse_cfg_raw = self._cfg("retrieval.sparse", {}) or {}
+            fusion_cfg_raw = self._cfg("retrieval.fusion", {}) or {}
+            graph_recall_cfg_raw = self._cfg("retrieval.search.graph_recall", {}) or {}
+            if not isinstance(sparse_cfg_raw, dict):
+                sparse_cfg_raw = {}
+            if not isinstance(fusion_cfg_raw, dict):
+                fusion_cfg_raw = {}
+            if not isinstance(graph_recall_cfg_raw, dict):
+                graph_recall_cfg_raw = {}
+
+            sparse_cfg = SparseBM25Config(**sparse_cfg_raw)
+            fusion_cfg = FusionConfig(**fusion_cfg_raw)
+            graph_recall_cfg = GraphRelationRecallConfig(**graph_recall_cfg_raw)
+            config = DualPathRetrieverConfig(
+                top_k_paragraphs=int(self._cfg("retrieval.top_k_paragraphs", 20)),
+                top_k_relations=int(self._cfg("retrieval.top_k_relations", 10)),
+                top_k_final=int(self._cfg("retrieval.top_k_final", 10)),
+                alpha=float(self._cfg("retrieval.alpha", 0.5)),
+                enable_ppr=bool(self._cfg("retrieval.enable_ppr", True)),
+                ppr_alpha=float(self._cfg("retrieval.ppr_alpha", 0.85)),
+                ppr_concurrency_limit=int(self._cfg("retrieval.ppr_concurrency_limit", 4)),
+                enable_parallel=bool(self._cfg("retrieval.enable_parallel", True)),
+                retrieval_strategy=RetrievalStrategy.DUAL_PATH,
+                debug=bool(self._cfg("advanced.debug", False)),
+                sparse=sparse_cfg,
+                fusion=fusion_cfg,
+                graph_recall=graph_recall_cfg,
+            )
+            return DualPathRetriever(
+                vector_store=self.vector_store,
+                graph_store=self.graph_store,
+                metadata_store=self.metadata_store,
+                embedding_manager=self.embedding_manager,
+                sparse_index=self.sparse_index,
+                config=config,
+            )
+        except Exception as e:
+            logger.warning(f"初始化人物画像检索器失败，将只使用关系证据: {e}")
+            return None
+
+    @staticmethod
+    def resolve_person_id(identifier: str) -> str:
+        """按 person_id 或姓名/别名解析 person_id。"""
+        if not identifier:
+            return ""
+        key = str(identifier).strip()
+        if not key:
+            return ""
+
+        if len(key) == 32 and all(ch in "0123456789abcdefABCDEF" for ch in key):
+            return key.lower()
+
+        try:
+            record = (
+                PersonInfo.select(PersonInfo.person_id)
+                .where((PersonInfo.person_name == key) | (PersonInfo.nickname == key))
+                .first()
+            )
+            if record and record.person_id:
+                return str(record.person_id)
+        except Exception:
+            pass
+
+        try:
+            record = (
+                PersonInfo.select(PersonInfo.person_id)
+                .where(PersonInfo.group_nick_name.contains(key))
+                .first()
+            )
+            if record and record.person_id:
+                return str(record.person_id)
+        except Exception:
+            pass
+
+        return ""
+
+    def _parse_group_nicks(self, raw_value: Any) -> List[str]:
+        if not raw_value:
+            return []
+        if isinstance(raw_value, list):
+            items = raw_value
+        else:
+            try:
+                items = json.loads(raw_value)
+            except Exception:
+                return []
+        names: List[str] = []
+        for item in items:
+            if isinstance(item, dict):
+                value = str(item.get("group_nick_name", "")).strip()
+                if value:
+                    names.append(value)
+            elif isinstance(item, str):
+                value = item.strip()
+                if value:
+                    names.append(value)
+        return names
+
+    def _parse_memory_traits(self, raw_value: Any) -> List[str]:
+        if not raw_value:
+            return []
+        try:
+            values = json.loads(raw_value) if isinstance(raw_value, str) else raw_value
+        except Exception:
+            return []
+        if not isinstance(values, list):
+            return []
+        traits: List[str] = []
+        for item in values:
+            text = str(item).strip()
+            if not text:
+                continue
+            if ":" in text:
+                parts = text.split(":")
+                if len(parts) >= 3:
+                    content = ":".join(parts[1:-1]).strip()
+                    if content:
+                        traits.append(content)
+                        continue
+            traits.append(text)
+        return traits[:10]
+
+    def get_person_aliases(self, person_id: str) -> Tuple[List[str], str, List[str]]:
+        """获取人物别名集合、主展示名、记忆特征。"""
+        aliases: List[str] = []
+        primary_name = ""
+        memory_traits: List[str] = []
+        if not person_id:
+            return aliases, primary_name, memory_traits
+        try:
+            record = PersonInfo.get_or_none(PersonInfo.person_id == person_id)
+            if not record:
+                return aliases, primary_name, memory_traits
+            person_name = str(getattr(record, "person_name", "") or "").strip()
+            nickname = str(getattr(record, "nickname", "") or "").strip()
+            group_nicks = self._parse_group_nicks(getattr(record, "group_nick_name", None))
+            memory_traits = self._parse_memory_traits(getattr(record, "memory_points", None))
+
+            primary_name = person_name or nickname or str(getattr(record, "user_id", "") or "").strip() or person_id
+
+            candidates = [person_name, nickname] + group_nicks
+            seen = set()
+            for item in candidates:
+                norm = str(item or "").strip()
+                if not norm or norm in seen:
+                    continue
+                seen.add(norm)
+                aliases.append(norm)
+        except Exception as e:
+            logger.warning(f"解析人物别名失败: person_id={person_id}, err={e}")
+        return aliases, primary_name, memory_traits
+
+    def _collect_relation_evidence(self, aliases: List[str], limit: int = 30) -> List[Dict[str, Any]]:
+        relation_by_hash: Dict[str, Dict[str, Any]] = {}
+        for alias in aliases:
+            for rel in self.metadata_store.get_relations(subject=alias):
+                h = str(rel.get("hash", ""))
+                if h:
+                    relation_by_hash[h] = rel
+            for rel in self.metadata_store.get_relations(object=alias):
+                h = str(rel.get("hash", ""))
+                if h:
+                    relation_by_hash[h] = rel
+
+        relations = list(relation_by_hash.values())
+        relations.sort(key=lambda item: float(item.get("confidence", 0.0)), reverse=True)
+        relations = relations[: max(1, int(limit))]
+
+        edges: List[Dict[str, Any]] = []
+        for rel in relations:
+            edges.append(
+                {
+                    "hash": str(rel.get("hash", "")),
+                    "subject": str(rel.get("subject", "")),
+                    "predicate": str(rel.get("predicate", "")),
+                    "object": str(rel.get("object", "")),
+                    "confidence": float(rel.get("confidence", 1.0) or 1.0),
+                }
+            )
+        return edges
+
+    async def _collect_vector_evidence(self, aliases: List[str], top_k: int = 12) -> List[Dict[str, Any]]:
+        alias_queries = [a for a in aliases if a]
+        if not alias_queries:
+            return []
+
+        if self.retriever is None:
+            # 回退：无检索器时只做简单内容匹配
+            fallback: List[Dict[str, Any]] = []
+            seen_hash = set()
+            for alias in alias_queries:
+                for para in self.metadata_store.search_paragraphs_by_content(alias)[: max(2, top_k // 2)]:
+                    h = str(para.get("hash", ""))
+                    if not h or h in seen_hash:
+                        continue
+                    seen_hash.add(h)
+                    fallback.append(
+                        {
+                            "hash": h,
+                            "type": "paragraph",
+                            "score": 0.0,
+                            "content": str(para.get("content", ""))[:180],
+                            "metadata": {},
+                        }
+                    )
+            return fallback[:top_k]
+
+        per_alias_top_k = max(2, int(top_k / max(1, len(alias_queries))))
+        seen_hash = set()
+        evidence: List[Dict[str, Any]] = []
+        for alias in alias_queries:
+            try:
+                results = await self.retriever.retrieve(alias, top_k=per_alias_top_k)
+            except Exception as e:
+                logger.warning(f"向量证据召回失败: alias={alias}, err={e}")
+                continue
+            for item in results:
+                h = str(getattr(item, "hash_value", "") or "")
+                if not h or h in seen_hash:
+                    continue
+                seen_hash.add(h)
+                evidence.append(
+                    {
+                        "hash": h,
+                        "type": str(getattr(item, "result_type", "")),
+                        "score": float(getattr(item, "score", 0.0) or 0.0),
+                        "content": str(getattr(item, "content", "") or "")[:220],
+                        "metadata": dict(getattr(item, "metadata", {}) or {}),
+                    }
+                )
+        evidence.sort(key=lambda x: x.get("score", 0.0), reverse=True)
+        return evidence[:top_k]
+
+    def _build_profile_text(
+        self,
+        person_id: str,
+        primary_name: str,
+        aliases: List[str],
+        relation_edges: List[Dict[str, Any]],
+        vector_evidence: List[Dict[str, Any]],
+        memory_traits: List[str],
+    ) -> str:
+        """基于证据构建画像文本（供 LLM 上下文注入）。"""
+        lines: List[str] = []
+        lines.append(f"人物ID: {person_id}")
+        if primary_name:
+            lines.append(f"主称呼: {primary_name}")
+        if aliases:
+            lines.append(f"别名: {', '.join(aliases[:8])}")
+        if memory_traits:
+            lines.append(f"记忆特征: {'; '.join(memory_traits[:6])}")
+
+        if relation_edges:
+            lines.append("关系证据:")
+            for rel in relation_edges[:6]:
+                s = rel.get("subject", "")
+                p = rel.get("predicate", "")
+                o = rel.get("object", "")
+                conf = float(rel.get("confidence", 0.0))
+                lines.append(f"- {s} {p} {o} (conf={conf:.2f})")
+
+        if vector_evidence:
+            lines.append("向量证据摘要:")
+            for item in vector_evidence[:4]:
+                content = str(item.get("content", "")).strip()
+                if content:
+                    lines.append(f"- {content}")
+
+        if len(lines) <= 2:
+            lines.append("暂无足够证据形成稳定画像。")
+
+        return "\n".join(lines)
+
+    @staticmethod
+    def _is_snapshot_stale(snapshot: Optional[Dict[str, Any]], ttl_seconds: float) -> bool:
+        if not snapshot:
+            return True
+        now = time.time()
+        expires_at = snapshot.get("expires_at")
+        if expires_at is not None:
+            try:
+                return now >= float(expires_at)
+            except Exception:
+                return True
+        updated_at = float(snapshot.get("updated_at") or 0.0)
+        return (now - updated_at) >= ttl_seconds
+
+    def _apply_manual_override(self, person_id: str, profile_payload: Dict[str, Any]) -> Dict[str, Any]:
+        """将手工覆盖并入画像结果（覆盖 profile_text，同时保留 auto_profile_text）。"""
+        payload = dict(profile_payload or {})
+        auto_text = str(payload.get("profile_text", "") or "")
+        payload["auto_profile_text"] = auto_text
+        payload["has_manual_override"] = False
+        payload["manual_override_text"] = ""
+        payload["override_updated_at"] = None
+        payload["override_updated_by"] = ""
+        payload["profile_source"] = "auto_snapshot"
+
+        if not person_id or self.metadata_store is None:
+            return payload
+
+        try:
+            override = self.metadata_store.get_person_profile_override(person_id)
+        except Exception as e:
+            logger.warning(f"读取人物画像手工覆盖失败: person_id={person_id}, err={e}")
+            return payload
+
+        if not override:
+            return payload
+
+        manual_text = str(override.get("override_text", "") or "").strip()
+        if not manual_text:
+            return payload
+
+        payload["has_manual_override"] = True
+        payload["manual_override_text"] = manual_text
+        payload["override_updated_at"] = override.get("updated_at")
+        payload["override_updated_by"] = str(override.get("updated_by", "") or "")
+        payload["profile_text"] = manual_text
+        payload["profile_source"] = "manual_override"
+        return payload
+
+    async def query_person_profile(
+        self,
+        person_id: str = "",
+        person_keyword: str = "",
+        top_k: int = 12,
+        ttl_seconds: float = 6 * 3600,
+        force_refresh: bool = False,
+        source_note: str = "",
+    ) -> Dict[str, Any]:
+        """查询或刷新人物画像。"""
+        pid = str(person_id or "").strip()
+        if not pid and person_keyword:
+            pid = self.resolve_person_id(person_keyword)
+
+        if not pid:
+            return {
+                "success": False,
+                "error": "person_id 无效，且未能通过别名解析",
+            }
+
+        latest = self.metadata_store.get_latest_person_profile_snapshot(pid)
+        if not force_refresh and not self._is_snapshot_stale(latest, ttl_seconds):
+            aliases, primary_name, _ = self.get_person_aliases(pid)
+            payload = {
+                "success": True,
+                "person_id": pid,
+                "person_name": primary_name,
+                "from_cache": True,
+                **(latest or {}),
+            }
+            if aliases and not payload.get("aliases"):
+                payload["aliases"] = aliases
+            return {
+                **self._apply_manual_override(pid, payload),
+            }
+
+        aliases, primary_name, memory_traits = self.get_person_aliases(pid)
+        if not aliases and person_keyword:
+            aliases = [person_keyword.strip()]
+            primary_name = person_keyword.strip()
+        relation_edges = self._collect_relation_evidence(aliases, limit=max(10, top_k * 2))
+        vector_evidence = await self._collect_vector_evidence(aliases, top_k=max(4, top_k))
+
+        evidence_ids = [
+            str(item.get("hash", ""))
+            for item in (relation_edges + vector_evidence)
+            if str(item.get("hash", "")).strip()
+        ]
+        dedup_ids: List[str] = []
+        seen = set()
+        for item in evidence_ids:
+            if item in seen:
+                continue
+            seen.add(item)
+            dedup_ids.append(item)
+
+        profile_text = self._build_profile_text(
+            person_id=pid,
+            primary_name=primary_name,
+            aliases=aliases,
+            relation_edges=relation_edges,
+            vector_evidence=vector_evidence,
+            memory_traits=memory_traits,
+        )
+
+        expires_at = time.time() + float(ttl_seconds) if ttl_seconds > 0 else None
+        snapshot = self.metadata_store.upsert_person_profile_snapshot(
+            person_id=pid,
+            profile_text=profile_text,
+            aliases=aliases,
+            relation_edges=relation_edges,
+            vector_evidence=vector_evidence,
+            evidence_ids=dedup_ids,
+            expires_at=expires_at,
+            source_note=source_note,
+        )
+        payload = {
+            "success": True,
+            "person_id": pid,
+            "person_name": primary_name,
+            "from_cache": False,
+            **snapshot,
+        }
+        return {
+            **self._apply_manual_override(pid, payload),
+        }
+
+    @staticmethod
+    def format_persona_profile_block(profile: Dict[str, Any]) -> str:
+        """格式化给 replyer 的注入块。"""
+        if not profile or not profile.get("success"):
+            return ""
+        text = str(profile.get("profile_text", "") or "").strip()
+        if not text:
+            return ""
+        return (
+            "【人物画像-内部参考】\n"
+            f"{text}\n"
+            "仅供内部推理，不要向用户逐字复述。"
+        )

plugins/A_memorix/core/utils/plugin_id_policy.py

@@ -0,0 +1,27 @@
+"""Plugin ID matching policy for A_Memorix."""
+
+from __future__ import annotations
+
+from typing import Any
+
+
+class PluginIdPolicy:
+    """Centralized plugin id normalization/matching policy."""
+
+    CANONICAL_ID = "a_memorix"
+
+    @classmethod
+    def normalize(cls, plugin_id: Any) -> str:
+        if not isinstance(plugin_id, str):
+            return ""
+        return plugin_id.strip().lower()
+
+    @classmethod
+    def is_target_plugin_id(cls, plugin_id: Any) -> bool:
+        normalized = cls.normalize(plugin_id)
+        if not normalized:
+            return False
+        if normalized == cls.CANONICAL_ID:
+            return True
+        return normalized.split(".")[-1] == cls.CANONICAL_ID
+

plugins/A_memorix/core/utils/quantization.py

@@ -0,0 +1,344 @@
+"""
+向量量化工具模块
+
+提供向量量化与反量化功能，用于压缩存储空间。
+"""
+
+import numpy as np
+from enum import Enum
+from typing import Tuple, Union
+
+from src.common.logger import get_logger
+
+logger = get_logger("A_Memorix.Quantization")
+
+
+class QuantizationType(Enum):
+    """量化类型枚举"""
+    FLOAT32 = "float32"  # 无量化
+    INT8 = "int8"        # 标量量化（8位整数）
+    PQ = "pq"            # 乘积量化（Product Quantization）
+
+
+def quantize_vector(
+    vector: np.ndarray,
+    quant_type: QuantizationType = QuantizationType.INT8,
+) -> Union[np.ndarray, Tuple[np.ndarray, np.ndarray]]:
+    """
+    量化向量
+
+    Args:
+        vector: 输入向量（float32）
+        quant_type: 量化类型
+
+    Returns:
+        量化后的向量：
+        - INT8: int8向量
+        - PQ: (编码向量, 聚类中心) 元组
+    """
+    if quant_type == QuantizationType.FLOAT32:
+        return vector.astype(np.float32)
+
+    elif quant_type == QuantizationType.INT8:
+        return _scalar_quantize_int8(vector)
+
+    elif quant_type == QuantizationType.PQ:
+        return _product_quantize(vector)
+
+    else:
+        raise ValueError(f"不支持的量化类型: {quant_type}")
+
+
+def dequantize_vector(
+    quantized_vector: Union[np.ndarray, Tuple[np.ndarray, np.ndarray]],
+    quant_type: QuantizationType = QuantizationType.INT8,
+    original_shape: Tuple[int, ...] = None,
+) -> np.ndarray:
+    """
+    反量化向量
+
+    Args:
+        quantized_vector: 量化后的向量
+        quant_type: 量化类型
+        original_shape: 原始向量形状（用于PQ）
+
+    Returns:
+        反量化后的向量（float32）
+    """
+    if quant_type == QuantizationType.FLOAT32:
+        return quantized_vector.astype(np.float32)
+
+    elif quant_type == QuantizationType.INT8:
+        return _scalar_dequantize_int8(quantized_vector)
+
+    elif quant_type == QuantizationType.PQ:
+        if not isinstance(quantized_vector, tuple):
+            raise ValueError("PQ反量化需要列表/元组格式: (codes, centroids)")
+        return _product_dequantize(quantized_vector[0], quantized_vector[1])
+
+    else:
+        raise ValueError(f"不支持的量化类型: {quant_type}")
+
+
+def _scalar_quantize_int8(vector: np.ndarray) -> np.ndarray:
+    """
+    标量量化：float32 -> int8
+
+    将向量归一化到 [0, 255] 范围，然后映射到 int8
+
+    Args:
+        vector: 输入向量
+
+    Returns:
+        量化后的 int8 向量
+    """
+    # 计算最小最大值
+    min_val = np.min(vector)
+    max_val = np.max(vector)
+
+    # 避免除零
+    if max_val == min_val:
+        return np.zeros_like(vector, dtype=np.int8)
+
+    # 归一化到 [0, 255]
+    normalized = (vector - min_val) / (max_val - min_val) * 255
+    
+    # 映射到 [-128, 127] 并转换为 int8
+    # np.round might return float, minus 128 then cast
+    quantized = np.round(normalized - 128.0).astype(np.int8)
+
+    # 存储归一化参数（用于反量化）
+    # 在实际存储中，这些参数需要单独保存
+    # 这里为了简单，我们使用一个全局字典来模拟
+    if not hasattr(_scalar_quantize_int8, "_params"):
+        _scalar_quantize_int8._params = {}
+
+    vector_id = id(vector)
+    _scalar_quantize_int8._params[vector_id] = (min_val, max_val)
+
+    return quantized
+
+
+def _scalar_dequantize_int8(quantized: np.ndarray) -> np.ndarray:
+    """
+    标量反量化：int8 -> float32
+
+    Args:
+        quantized: 量化后的 int8 向量
+
+    Returns:
+        反量化后的 float32 向量
+    """
+    # 计算归一化参数（如果提供了）
+    # 在实际应用中，min_val 和 max_val 应该被保存
+    if not hasattr(_scalar_dequantize_int8, "_params"):
+        # 默认假设范围是 [-1, 1]
+        return (quantized.astype(np.float32) + 128.0) / 255.0 * 2.0 - 1.0
+
+    # 尝试查找参数 (这里只是演示逻辑，实际应从存储中读取)
+    # return (quantized.astype(np.float32) + 128.0) / 255.0 * (max - min) + min
+    return (quantized.astype(np.float32) + 128.0) / 255.0
+
+
+def quantize_matrix(
+    matrix: np.ndarray,
+    quant_type: QuantizationType = QuantizationType.INT8,
+) -> Union[np.ndarray, Tuple[np.ndarray, np.ndarray]]:
+    """
+    量化矩阵（批量量化向量）
+
+    Args:
+        matrix: 输入矩阵（N x D，每行是一个向量）
+        quant_type: 量化类型
+
+    Returns:
+        量化后的矩阵
+    """
+    if quant_type == QuantizationType.FLOAT32:
+        return matrix.astype(np.float32)
+
+    elif quant_type == QuantizationType.INT8:
+        # 对整个矩阵进行全局归一化
+        min_val = np.min(matrix)
+        max_val = np.max(matrix)
+
+        if max_val == min_val:
+            return np.zeros_like(matrix, dtype=np.int8)
+
+        # 归一化到 [0, 255]
+        normalized = (matrix - min_val) / (max_val - min_val) * 255
+        quantized = np.round(normalized).astype(np.int8)
+
+        return quantized
+
+    else:
+        raise ValueError(f"不支持的量化类型: {quant_type}")
+
+
+def dequantize_matrix(
+    quantized_matrix: np.ndarray,
+    quant_type: QuantizationType = QuantizationType.INT8,
+    min_val: float = None,
+    max_val: float = None,
+) -> np.ndarray:
+    """
+    反量化矩阵
+
+    Args:
+        quantized_matrix: 量化后的矩阵
+        quant_type: 量化类型
+        min_val: 归一化最小值（int8反量化需要）
+        max_val: 归一化最大值（int8反量化需要）
+
+    Returns:
+        反量化后的矩阵
+    """
+    if quant_type == QuantizationType.FLOAT32:
+        return quantized_matrix.astype(np.float32)
+
+    elif quant_type == QuantizationType.INT8:
+        # 使用提供的归一化参数反量化
+        if min_val is None or max_val is None:
+            # 默认假设范围是 [0, 255] -> [-1, 1]
+            return quantized_matrix.astype(np.float32) / 127.0
+        else:
+            # 恢复到原始范围
+            normalized = quantized_matrix.astype(np.float32) / 255.0
+            return normalized * (max_val - min_val) + min_val
+
+    else:
+        raise ValueError(f"不支持的量化类型: {quant_type}")
+
+
+def estimate_memory_reduction(
+    num_vectors: int,
+    dimension: int,
+    from_type: QuantizationType,
+    to_type: QuantizationType,
+) -> Tuple[float, float]:
+    """
+    估算内存节省量
+
+    Args:
+        num_vectors: 向量数量
+        dimension: 向量维度
+        from_type: 原始量化类型
+        to_type: 目标量化类型
+
+    Returns:
+        (原始大小MB, 量化后大小MB, 节省比例)
+    """
+    # 计算每个向量占用的字节数
+    bytes_per_element = {
+        QuantizationType.FLOAT32: 4,
+        QuantizationType.INT8: 1,
+        QuantizationType.PQ: 0.25,  # 假设压缩到1/4
+    }
+
+    original_bytes = num_vectors * dimension * bytes_per_element[from_type]
+    quantized_bytes = num_vectors * dimension * bytes_per_element[to_type]
+
+    original_mb = original_bytes / 1024 / 1024
+    quantized_mb = quantized_bytes / 1024 / 1024
+    reduction_ratio = (original_bytes - quantized_bytes) / original_bytes
+
+    return original_mb, quantized_mb, reduction_ratio
+
+
+def estimate_compression_stats(
+    num_vectors: int,
+    dimension: int,
+    quant_type: QuantizationType,
+) -> dict:
+    """
+    估算压缩统计信息
+
+    Args:
+        num_vectors: 向量数量
+        dimension: 向量维度
+        quant_type: 量化类型
+
+    Returns:
+        统计信息字典
+    """
+    original_mb, quantized_mb, ratio = estimate_memory_reduction(
+        num_vectors, dimension, QuantizationType.FLOAT32, quant_type
+    )
+
+    return {
+        "num_vectors": num_vectors,
+        "dimension": dimension,
+        "quantization_type": quant_type.value,
+        "original_size_mb": round(original_mb, 2),
+        "quantized_size_mb": round(quantized_mb, 2),
+        "saved_mb": round(original_mb - quantized_mb, 2),
+        "compression_ratio": round(ratio * 100, 2),
+    }
+
+
+def _product_quantize(
+    vector: np.ndarray, m: int = 8, k: int = 256
+) -> Tuple[np.ndarray, np.ndarray]:
+    """
+    乘积量化 (PQ) 简化实现
+
+    Args:
+        vector: 输入向量 (D,)
+        m: 子空间数量
+        k: 每个子空间的聚类中心数
+
+    Returns:
+        (编码后的向量, 聚类中心)
+    """
+    d = vector.shape[0]
+    if d % m != 0:
+        raise ValueError(f"维度 {d} 必须能被子空间数量 {m} 整除")
+
+    ds = d // m  # 子空间维度
+    codes = np.zeros(m, dtype=np.uint8)
+    centroids = np.zeros((m, k, ds), dtype=np.float32)
+
+    # 这里采用一种简化的 PQ：不进行 K-means 训练，
+    # 而是预定一些量化点或针对单向量的微型聚类（实际应用中应离线训练）
+    # 为了演示，我们直接将子空间切分为 k 份进行量化
+    for i in range(m):
+        sub_vec = vector[i * ds : (i + 1) * ds]
+        # 简化：假定每个子空间的取值范围并划分
+        # 实际 PQ 应使用 k-means 产生的 centroids
+        # 这里为演示创建一个随机 codebook 并找到最接近的核心
+        sub_min, sub_max = np.min(sub_vec), np.max(sub_vec)
+        if sub_max == sub_min:
+            linspace = np.zeros(k)
+        else:
+            linspace = np.linspace(sub_min, sub_max, k)
+        
+        for j in range(k):
+             centroids[i, j, :] = linspace[j]
+        
+        # 编码：这里简化为取子空间均值找最接近的 centroid
+        sub_mean = np.mean(sub_vec)
+        code = np.argmin(np.abs(linspace - sub_mean))
+        codes[i] = code
+
+    return codes, centroids
+
+
+def _product_dequantize(codes: np.ndarray, centroids: np.ndarray) -> np.ndarray:
+    """
+    PQ 反量化
+
+    Args:
+        codes: 编码向量 (M,)
+        centroids: 聚类中心 (M, K, DS)
+
+    Returns:
+        恢复后的向量 (D,)
+    """
+    m, k, ds = centroids.shape
+    vector = np.zeros(m * ds, dtype=np.float32)
+
+    for i in range(m):
+        code = codes[i]
+        vector[i * ds : (i + 1) * ds] = centroids[i, code, :]
+
+    return vector

plugins/A_memorix/core/utils/relation_query.py

@@ -0,0 +1,121 @@
+"""关系查询规格解析工具。"""
+
+from __future__ import annotations
+
+import re
+from dataclasses import dataclass
+from typing import Optional
+
+
+@dataclass
+class RelationQuerySpec:
+    raw: str
+    is_structured: bool
+    subject: Optional[str]
+    predicate: Optional[str]
+    object: Optional[str]
+    error: Optional[str] = None
+
+
+_NATURAL_LANGUAGE_PATTERN = re.compile(
+    r"(^\s*(what|who|which|how|why|when|where)\b|"
+    r"\?|？|"
+    r"\b(relation|related|between)\b|"
+    r"(什么关系|有哪些关系|之间|关联))",
+    re.IGNORECASE,
+)
+
+
+def _looks_like_natural_language(raw: str) -> bool:
+    text = str(raw or "").strip()
+    if not text:
+        return False
+    return _NATURAL_LANGUAGE_PATTERN.search(text) is not None
+
+
+def parse_relation_query_spec(relation_spec: str) -> RelationQuerySpec:
+    raw = str(relation_spec or "").strip()
+    if not raw:
+        return RelationQuerySpec(
+            raw=raw,
+            is_structured=False,
+            subject=None,
+            predicate=None,
+            object=None,
+            error="empty",
+        )
+
+    if "|" in raw:
+        parts = [p.strip() for p in raw.split("|")]
+        if len(parts) < 2:
+            return RelationQuerySpec(
+                raw=raw,
+                is_structured=True,
+                subject=None,
+                predicate=None,
+                object=None,
+                error="invalid_pipe_format",
+            )
+        return RelationQuerySpec(
+            raw=raw,
+            is_structured=True,
+            subject=parts[0] or None,
+            predicate=parts[1] or None,
+            object=parts[2] if len(parts) > 2 and parts[2] else None,
+        )
+
+    if "->" in raw:
+        parts = [p.strip() for p in raw.split("->") if p.strip()]
+        if len(parts) >= 3:
+            return RelationQuerySpec(
+                raw=raw,
+                is_structured=True,
+                subject=parts[0],
+                predicate=parts[1],
+                object=parts[2],
+            )
+        if len(parts) == 2:
+            return RelationQuerySpec(
+                raw=raw,
+                is_structured=True,
+                subject=parts[0],
+                predicate=None,
+                object=parts[1],
+            )
+        return RelationQuerySpec(
+            raw=raw,
+            is_structured=True,
+            subject=None,
+            predicate=None,
+            object=None,
+            error="invalid_arrow_format",
+        )
+
+    if _looks_like_natural_language(raw):
+        return RelationQuerySpec(
+            raw=raw,
+            is_structured=False,
+            subject=None,
+            predicate=None,
+            object=None,
+        )
+
+    # 仅保留低歧义的紧凑三元组作为兼容语法，例如 "Alice likes Apple"。
+    # 两词形式过于模糊，不再视为结构化关系查询。
+    parts = raw.split()
+    if len(parts) == 3:
+        return RelationQuerySpec(
+            raw=raw,
+            is_structured=True,
+            subject=parts[0],
+            predicate=parts[1],
+            object=parts[2],
+        )
+
+    return RelationQuerySpec(
+        raw=raw,
+        is_structured=False,
+        subject=None,
+        predicate=None,
+        object=None,
+    )

plugins/A_memorix/core/utils/relation_write_service.py

@@ -0,0 +1,164 @@
+"""
+统一关系写入与关系向量化服务。
+
+规则：
+1. 元数据是主数据源，向量是从索引。
+2. 关系先写 metadata，再写向量。
+3. 向量失败不回滚 metadata，依赖状态机与回填任务修复。
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Any, Dict, Optional
+
+from src.common.logger import get_logger
+
+
+logger = get_logger("A_Memorix.RelationWriteService")
+
+
+@dataclass
+class RelationWriteResult:
+    hash_value: str
+    vector_written: bool
+    vector_already_exists: bool
+    vector_state: str
+
+
+class RelationWriteService:
+    """关系写入收口服务。"""
+
+    ERROR_MAX_LEN = 500
+
+    def __init__(
+        self,
+        metadata_store: Any,
+        graph_store: Any,
+        vector_store: Any,
+        embedding_manager: Any,
+    ):
+        self.metadata_store = metadata_store
+        self.graph_store = graph_store
+        self.vector_store = vector_store
+        self.embedding_manager = embedding_manager
+
+    @staticmethod
+    def build_relation_vector_text(subject: str, predicate: str, obj: str) -> str:
+        s = str(subject or "").strip()
+        p = str(predicate or "").strip()
+        o = str(obj or "").strip()
+        # 双表达：兼容关键词检索与自然语言问句
+        return f"{s} {p} {o}\n{s}和{o}的关系是{p}"
+
+    async def ensure_relation_vector(
+        self,
+        hash_value: str,
+        subject: str,
+        predicate: str,
+        obj: str,
+        *,
+        max_error_len: int = ERROR_MAX_LEN,
+    ) -> RelationWriteResult:
+        """
+        为已有关系确保向量存在并更新状态。
+        """
+        if hash_value in self.vector_store:
+            self.metadata_store.set_relation_vector_state(hash_value, "ready")
+            return RelationWriteResult(
+                hash_value=hash_value,
+                vector_written=False,
+                vector_already_exists=True,
+                vector_state="ready",
+            )
+
+        self.metadata_store.set_relation_vector_state(hash_value, "pending")
+        try:
+            vector_text = self.build_relation_vector_text(subject, predicate, obj)
+            embedding = await self.embedding_manager.encode(vector_text)
+            self.vector_store.add(
+                vectors=embedding.reshape(1, -1),
+                ids=[hash_value],
+            )
+            self.metadata_store.set_relation_vector_state(hash_value, "ready")
+            logger.info(
+                "metric.relation_vector_write_success=1 metric.relation_vector_write_success_count=1 hash=%s",
+                hash_value[:16],
+            )
+            return RelationWriteResult(
+                hash_value=hash_value,
+                vector_written=True,
+                vector_already_exists=False,
+                vector_state="ready",
+            )
+        except ValueError:
+            # 向量已存在冲突，按成功处理
+            self.metadata_store.set_relation_vector_state(hash_value, "ready")
+            return RelationWriteResult(
+                hash_value=hash_value,
+                vector_written=False,
+                vector_already_exists=True,
+                vector_state="ready",
+            )
+        except Exception as e:
+            err = str(e)[:max_error_len]
+            self.metadata_store.set_relation_vector_state(
+                hash_value,
+                "failed",
+                error=err,
+                bump_retry=True,
+            )
+            logger.warning(
+                "metric.relation_vector_write_fail=1 metric.relation_vector_write_fail_count=1 hash=%s err=%s",
+                hash_value[:16],
+                err,
+            )
+            return RelationWriteResult(
+                hash_value=hash_value,
+                vector_written=False,
+                vector_already_exists=False,
+                vector_state="failed",
+            )
+
+    async def upsert_relation_with_vector(
+        self,
+        subject: str,
+        predicate: str,
+        obj: str,
+        confidence: float = 1.0,
+        source_paragraph: str = "",
+        metadata: Optional[Dict[str, Any]] = None,
+        *,
+        write_vector: bool = True,
+    ) -> RelationWriteResult:
+        """
+        统一关系写入：
+        1) 写 metadata relation
+        2) 写 graph edge relation_hash
+        3) 按需写 relation vector
+        """
+        rel_hash = self.metadata_store.add_relation(
+            subject=subject,
+            predicate=predicate,
+            obj=obj,
+            confidence=confidence,
+            source_paragraph=source_paragraph,
+            metadata=metadata or {},
+        )
+        self.graph_store.add_edges([(subject, obj)], relation_hashes=[rel_hash])
+
+        if not write_vector:
+            self.metadata_store.set_relation_vector_state(rel_hash, "none")
+            return RelationWriteResult(
+                hash_value=rel_hash,
+                vector_written=False,
+                vector_already_exists=False,
+                vector_state="none",
+            )
+
+        return await self.ensure_relation_vector(
+            hash_value=rel_hash,
+            subject=subject,
+            predicate=predicate,
+            obj=obj,
+        )

plugins/A_memorix/core/utils/runtime_self_check.py

@@ -0,0 +1,197 @@
+"""Runtime self-check helpers for A_Memorix."""
+
+from __future__ import annotations
+
+import time
+from typing import Any, Dict, Optional
+
+import numpy as np
+
+from src.common.logger import get_logger
+
+logger = get_logger("A_Memorix.RuntimeSelfCheck")
+
+_DEFAULT_SAMPLE_TEXT = "A_Memorix runtime self check"
+
+
+def _safe_int(value: Any, default: int = 0) -> int:
+    try:
+        return int(value)
+    except Exception:
+        return int(default)
+
+
+def _get_config_value(config: Any, key: str, default: Any = None) -> Any:
+    getter = getattr(config, "get_config", None)
+    if callable(getter):
+        return getter(key, default)
+
+    current: Any = config
+    for part in key.split("."):
+        if isinstance(current, dict) and part in current:
+            current = current[part]
+        else:
+            return default
+    return current
+
+
+def _build_report(
+    *,
+    ok: bool,
+    code: str,
+    message: str,
+    configured_dimension: int,
+    vector_store_dimension: int,
+    detected_dimension: int,
+    encoded_dimension: int,
+    elapsed_ms: float,
+    sample_text: str,
+) -> Dict[str, Any]:
+    return {
+        "ok": bool(ok),
+        "code": str(code or "").strip(),
+        "message": str(message or "").strip(),
+        "configured_dimension": int(configured_dimension),
+        "vector_store_dimension": int(vector_store_dimension),
+        "detected_dimension": int(detected_dimension),
+        "encoded_dimension": int(encoded_dimension),
+        "elapsed_ms": float(elapsed_ms),
+        "sample_text": str(sample_text or ""),
+        "checked_at": time.time(),
+    }
+
+
+async def run_embedding_runtime_self_check(
+    *,
+    config: Any,
+    vector_store: Optional[Any],
+    embedding_manager: Optional[Any],
+    sample_text: str = _DEFAULT_SAMPLE_TEXT,
+) -> Dict[str, Any]:
+    """Probe the real embedding path and compare dimensions with runtime storage."""
+    configured_dimension = _safe_int(_get_config_value(config, "embedding.dimension", 0), 0)
+    vector_store_dimension = _safe_int(getattr(vector_store, "dimension", 0), 0)
+
+    if vector_store is None or embedding_manager is None:
+        return _build_report(
+            ok=False,
+            code="runtime_components_missing",
+            message="vector_store 或 embedding_manager 未初始化",
+            configured_dimension=configured_dimension,
+            vector_store_dimension=vector_store_dimension,
+            detected_dimension=0,
+            encoded_dimension=0,
+            elapsed_ms=0.0,
+            sample_text=sample_text,
+        )
+
+    start = time.perf_counter()
+    detected_dimension = 0
+    encoded_dimension = 0
+    try:
+        detected_dimension = _safe_int(await embedding_manager._detect_dimension(), 0)
+        encoded = await embedding_manager.encode(sample_text)
+        if isinstance(encoded, np.ndarray):
+            encoded_dimension = int(encoded.shape[0]) if encoded.ndim == 1 else int(encoded.shape[-1])
+        else:
+            encoded_dimension = len(encoded) if encoded is not None else 0
+    except Exception as exc:
+        elapsed_ms = (time.perf_counter() - start) * 1000.0
+        logger.warning("embedding runtime self-check failed: %s", exc)
+        return _build_report(
+            ok=False,
+            code="embedding_probe_failed",
+            message=f"embedding probe failed: {exc}",
+            configured_dimension=configured_dimension,
+            vector_store_dimension=vector_store_dimension,
+            detected_dimension=detected_dimension,
+            encoded_dimension=encoded_dimension,
+            elapsed_ms=elapsed_ms,
+            sample_text=sample_text,
+        )
+
+    elapsed_ms = (time.perf_counter() - start) * 1000.0
+    expected_dimension = vector_store_dimension or configured_dimension or detected_dimension
+    if expected_dimension <= 0:
+        return _build_report(
+            ok=False,
+            code="invalid_expected_dimension",
+            message="无法确定期望 embedding 维度",
+            configured_dimension=configured_dimension,
+            vector_store_dimension=vector_store_dimension,
+            detected_dimension=detected_dimension,
+            encoded_dimension=encoded_dimension,
+            elapsed_ms=elapsed_ms,
+            sample_text=sample_text,
+        )
+
+    if encoded_dimension != expected_dimension:
+        msg = (
+            "embedding 真实输出维度与当前向量存储不一致: "
+            f"expected={expected_dimension}, encoded={encoded_dimension}"
+        )
+        logger.error(msg)
+        return _build_report(
+            ok=False,
+            code="embedding_dimension_mismatch",
+            message=msg,
+            configured_dimension=configured_dimension,
+            vector_store_dimension=vector_store_dimension,
+            detected_dimension=detected_dimension,
+            encoded_dimension=encoded_dimension,
+            elapsed_ms=elapsed_ms,
+            sample_text=sample_text,
+        )
+
+    return _build_report(
+        ok=True,
+        code="ok",
+        message="embedding runtime self-check passed",
+        configured_dimension=configured_dimension,
+        vector_store_dimension=vector_store_dimension,
+        detected_dimension=detected_dimension,
+        encoded_dimension=encoded_dimension,
+        elapsed_ms=elapsed_ms,
+        sample_text=sample_text,
+    )
+
+
+async def ensure_runtime_self_check(
+    plugin_or_config: Any,
+    *,
+    force: bool = False,
+    sample_text: str = _DEFAULT_SAMPLE_TEXT,
+) -> Dict[str, Any]:
+    """Run or reuse cached runtime self-check report."""
+    if plugin_or_config is None:
+        return _build_report(
+            ok=False,
+            code="missing_plugin_or_config",
+            message="plugin/config unavailable",
+            configured_dimension=0,
+            vector_store_dimension=0,
+            detected_dimension=0,
+            encoded_dimension=0,
+            elapsed_ms=0.0,
+            sample_text=sample_text,
+        )
+
+    cache = getattr(plugin_or_config, "_runtime_self_check_report", None)
+    if isinstance(cache, dict) and cache and not force:
+        return cache
+
+    report = await run_embedding_runtime_self_check(
+        config=getattr(plugin_or_config, "config", plugin_or_config),
+        vector_store=getattr(plugin_or_config, "vector_store", None)
+        if not isinstance(plugin_or_config, dict)
+        else plugin_or_config.get("vector_store"),
+        embedding_manager=getattr(plugin_or_config, "embedding_manager", None)
+        if not isinstance(plugin_or_config, dict)
+        else plugin_or_config.get("embedding_manager"),
+        sample_text=sample_text,
+    )
+    try:
+        setattr(plugin_or_config, "_runtime_self_check_report", report)
+    except Exception:
+        pass
+    return report

plugins/A_memorix/core/utils/search_postprocess.py

@@ -0,0 +1,90 @@
+"""Post-processing helpers for unified search execution."""
+
+from __future__ import annotations
+
+from typing import Any, List, Tuple
+
+from .path_fallback_service import find_paths_from_query, to_retrieval_results
+
+
+def apply_safe_content_dedup(results: List[Any]) -> Tuple[List[Any], int]:
+    """Deduplicate results by hash/content while preserving at least one entry."""
+    if not results:
+        return [], 0
+
+    unique_results: List[Any] = []
+    seen_hashes = set()
+    seen_contents = set()
+
+    for item in results:
+        content = str(getattr(item, "content", "") or "").strip()
+        if not content:
+            continue
+
+        hash_value = str(getattr(item, "hash_value", "") or "").strip() or str(hash(content))
+        if hash_value in seen_hashes:
+            continue
+
+        is_dup = False
+        for seen in seen_contents:
+            if content in seen or seen in content:
+                is_dup = True
+                break
+        if is_dup:
+            continue
+
+        seen_hashes.add(hash_value)
+        seen_contents.add(content)
+        unique_results.append(item)
+
+    if not unique_results:
+        unique_results.append(results[0])
+
+    removed_count = max(0, len(results) - len(unique_results))
+    return unique_results, removed_count
+
+
+def maybe_apply_smart_path_fallback(
+    *,
+    query: str,
+    results: List[Any],
+    graph_store: Any,
+    metadata_store: Any,
+    enabled: bool,
+    threshold: float,
+    max_depth: int = 3,
+    max_paths: int = 5,
+) -> Tuple[List[Any], bool, int]:
+    """Append indirect relation paths when semantic results are weak."""
+    if not enabled or not str(query or "").strip():
+        return results, False, 0
+    if graph_store is None or metadata_store is None:
+        return results, False, 0
+
+    max_score = 0.0
+    if results:
+        try:
+            max_score = float(getattr(results[0], "score", 0.0) or 0.0)
+        except Exception:
+            max_score = 0.0
+
+    if max_score >= float(threshold):
+        return results, False, 0
+
+    paths = find_paths_from_query(
+        query=query,
+        graph_store=graph_store,
+        metadata_store=metadata_store,
+        max_depth=max_depth,
+        max_paths=max_paths,
+    )
+    if not paths:
+        return results, False, 0
+
+    path_results = to_retrieval_results(paths)
+    if not path_results:
+        return results, False, 0
+
+    merged = list(path_results) + list(results)
+    return merged, True, len(path_results)
+

plugins/A_memorix/core/utils/time_parser.py

@@ -0,0 +1,170 @@
+"""
+时间解析工具。
+
+约束：
+1. 查询参数（Action/Command/Tool）仅接受结构化绝对时间：
+   - YYYY/MM/DD
+   - YYYY/MM/DD HH:mm
+2. 入库时允许更宽松格式（含时间戳、YYYY-MM-DD 等）。
+"""
+
+from __future__ import annotations
+
+import re
+from datetime import datetime
+from typing import Any, Dict, Optional, Tuple
+
+
+_QUERY_DATE_RE = re.compile(r"^\d{4}/\d{2}/\d{2}$")
+_QUERY_MINUTE_RE = re.compile(r"^\d{4}/\d{2}/\d{2} \d{2}:\d{2}$")
+_NUMERIC_RE = re.compile(r"^-?\d+(?:\.\d+)?$")
+
+_INGEST_FORMATS = [
+    "%Y/%m/%d %H:%M:%S",
+    "%Y/%m/%d %H:%M",
+    "%Y-%m-%d %H:%M:%S",
+    "%Y-%m-%d %H:%M",
+    "%Y-%m-%dT%H:%M:%S",
+    "%Y-%m-%dT%H:%M",
+    "%Y/%m/%d",
+    "%Y-%m-%d",
+]
+
+_INGEST_DATE_FORMATS = {"%Y/%m/%d", "%Y-%m-%d"}
+
+
+def parse_query_datetime_to_timestamp(value: str, is_end: bool = False) -> float:
+    """解析查询时间，仅支持 YYYY/MM/DD 或 YYYY/MM/DD HH:mm。"""
+    text = str(value).strip()
+    if not text:
+        raise ValueError("时间不能为空")
+
+    if _QUERY_DATE_RE.fullmatch(text):
+        dt = datetime.strptime(text, "%Y/%m/%d")
+        if is_end:
+            dt = dt.replace(hour=23, minute=59, second=0, microsecond=0)
+        return dt.timestamp()
+
+    if _QUERY_MINUTE_RE.fullmatch(text):
+        dt = datetime.strptime(text, "%Y/%m/%d %H:%M")
+        return dt.timestamp()
+
+    raise ValueError(
+        f"时间格式错误: {text}。仅支持 YYYY/MM/DD 或 YYYY/MM/DD HH:mm"
+    )
+
+
+def parse_query_time_range(
+    time_from: Optional[str],
+    time_to: Optional[str],
+) -> Tuple[Optional[float], Optional[float]]:
+    """解析查询窗口并验证区间。"""
+    ts_from = (
+        parse_query_datetime_to_timestamp(time_from, is_end=False)
+        if time_from
+        else None
+    )
+    ts_to = (
+        parse_query_datetime_to_timestamp(time_to, is_end=True)
+        if time_to
+        else None
+    )
+
+    if ts_from is not None and ts_to is not None and ts_from > ts_to:
+        raise ValueError("time_from 不能晚于 time_to")
+
+    return ts_from, ts_to
+
+
+def parse_ingest_datetime_to_timestamp(
+    value: Any,
+    is_end: bool = False,
+) -> Optional[float]:
+    """解析入库时间，允许 timestamp/常见字符串格式。"""
+    if value is None:
+        return None
+
+    if isinstance(value, (int, float)):
+        return float(value)
+
+    text = str(value).strip()
+    if not text:
+        return None
+
+    if _NUMERIC_RE.fullmatch(text):
+        return float(text)
+
+    for fmt in _INGEST_FORMATS:
+        try:
+            dt = datetime.strptime(text, fmt)
+            if fmt in _INGEST_DATE_FORMATS and is_end:
+                dt = dt.replace(hour=23, minute=59, second=0, microsecond=0)
+            return dt.timestamp()
+        except ValueError:
+            continue
+
+    raise ValueError(f"无法解析时间: {text}")
+
+
+def normalize_time_meta(time_meta: Optional[Dict[str, Any]]) -> Dict[str, Any]:
+    """归一化 time_meta 到存储层字段。"""
+    if not time_meta:
+        return {}
+
+    normalized: Dict[str, Any] = {}
+
+    event_time = parse_ingest_datetime_to_timestamp(time_meta.get("event_time"))
+    event_start = parse_ingest_datetime_to_timestamp(
+        time_meta.get("event_time_start"),
+        is_end=False,
+    )
+    event_end = parse_ingest_datetime_to_timestamp(
+        time_meta.get("event_time_end"),
+        is_end=True,
+    )
+
+    time_range = time_meta.get("time_range")
+    if (
+        isinstance(time_range, (list, tuple))
+        and len(time_range) == 2
+    ):
+        if event_start is None:
+            event_start = parse_ingest_datetime_to_timestamp(time_range[0], is_end=False)
+        if event_end is None:
+            event_end = parse_ingest_datetime_to_timestamp(time_range[1], is_end=True)
+
+    if event_start is not None and event_end is not None and event_start > event_end:
+        raise ValueError("event_time_start 不能晚于 event_time_end")
+
+    if event_time is not None:
+        normalized["event_time"] = event_time
+    if event_start is not None:
+        normalized["event_time_start"] = event_start
+    if event_end is not None:
+        normalized["event_time_end"] = event_end
+
+    granularity = time_meta.get("time_granularity")
+    if granularity:
+        normalized["time_granularity"] = str(granularity)
+    else:
+        raw_time_values = [
+            time_meta.get("event_time"),
+            time_meta.get("event_time_start"),
+            time_meta.get("event_time_end"),
+        ]
+        has_minute = any(isinstance(v, str) and ":" in v for v in raw_time_values if v is not None)
+        normalized["time_granularity"] = "minute" if has_minute else "day"
+
+    confidence = time_meta.get("time_confidence")
+    if confidence is not None:
+        normalized["time_confidence"] = float(confidence)
+
+    return normalized
+
+
+def format_timestamp(ts: Optional[float]) -> Optional[str]:
+    """将 timestamp 格式化为 YYYY/MM/DD HH:mm。"""
+    if ts is None:
+        return None
+    return datetime.fromtimestamp(ts).strftime("%Y/%m/%d %H:%M")
+

plugins/A_memorix/plugin.py

@@ -0,0 +1,207 @@
+"""A_Memorix SDK plugin entry."""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+
+from maibot_sdk import MaiBotPlugin, Tool
+from maibot_sdk.types import ToolParameterInfo, ToolParamType
+
+from A_memorix.core.runtime.sdk_memory_kernel import KernelSearchRequest, SDKMemoryKernel
+
+
+def _tool_param(name: str, param_type: ToolParamType, description: str, required: bool) -> ToolParameterInfo:
+    return ToolParameterInfo(name=name, param_type=param_type, description=description, required=required)
+
+
+class AMemorixPlugin(MaiBotPlugin):
+    def __init__(self) -> None:
+        super().__init__()
+        self._plugin_root = Path(__file__).resolve().parent
+        self._plugin_config: Dict[str, Any] = {}
+        self._kernel: Optional[SDKMemoryKernel] = None
+
+    def set_plugin_config(self, config: Dict[str, Any]) -> None:
+        self._plugin_config = config or {}
+        if self._kernel is not None:
+            self._kernel.close()
+            self._kernel = None
+
+    async def on_load(self):
+        await self._get_kernel()
+
+    async def on_unload(self):
+        if self._kernel is not None:
+            self._kernel.close()
+            self._kernel = None
+
+    async def _get_kernel(self) -> SDKMemoryKernel:
+        if self._kernel is None:
+            self._kernel = SDKMemoryKernel(plugin_root=self._plugin_root, config=self._plugin_config)
+            await self._kernel.initialize()
+        return self._kernel
+
+    @Tool(
+        "search_memory",
+        description="搜索长期记忆",
+        parameters=[
+            _tool_param("query", ToolParamType.STRING, "查询文本", False),
+            _tool_param("limit", ToolParamType.INTEGER, "返回条数", False),
+            _tool_param("mode", ToolParamType.STRING, "search/time/hybrid/episode/aggregate", False),
+            _tool_param("chat_id", ToolParamType.STRING, "聊天流 ID", False),
+            _tool_param("person_id", ToolParamType.STRING, "人物 ID", False),
+            _tool_param("time_start", ToolParamType.FLOAT, "起始时间戳", False),
+            _tool_param("time_end", ToolParamType.FLOAT, "结束时间戳", False),
+        ],
+    )
+    async def handle_search_memory(
+        self,
+        query: str = "",
+        limit: int = 5,
+        mode: str = "hybrid",
+        chat_id: str = "",
+        person_id: str = "",
+        time_start: float | None = None,
+        time_end: float | None = None,
+        **kwargs,
+    ):
+        _ = kwargs
+        kernel = await self._get_kernel()
+        return await kernel.search_memory(
+            KernelSearchRequest(
+                query=query,
+                limit=limit,
+                mode=mode,
+                chat_id=chat_id,
+                person_id=person_id,
+                time_start=time_start,
+                time_end=time_end,
+            )
+        )
+
+    @Tool(
+        "ingest_summary",
+        description="写入聊天摘要到长期记忆",
+        parameters=[
+            _tool_param("external_id", ToolParamType.STRING, "外部幂等 ID", True),
+            _tool_param("chat_id", ToolParamType.STRING, "聊天流 ID", True),
+            _tool_param("text", ToolParamType.STRING, "摘要文本", True),
+            _tool_param("time_start", ToolParamType.FLOAT, "起始时间戳", False),
+            _tool_param("time_end", ToolParamType.FLOAT, "结束时间戳", False),
+        ],
+    )
+    async def handle_ingest_summary(
+        self,
+        external_id: str,
+        chat_id: str,
+        text: str,
+        participants: Optional[List[str]] = None,
+        time_start: float | None = None,
+        time_end: float | None = None,
+        tags: Optional[List[str]] = None,
+        metadata: Optional[Dict[str, Any]] = None,
+        **kwargs,
+    ):
+        _ = kwargs
+        kernel = await self._get_kernel()
+        return await kernel.ingest_summary(
+            external_id=external_id,
+            chat_id=chat_id,
+            text=text,
+            participants=participants,
+            time_start=time_start,
+            time_end=time_end,
+            tags=tags,
+            metadata=metadata,
+        )
+
+    @Tool(
+        "ingest_text",
+        description="写入普通长期记忆文本",
+        parameters=[
+            _tool_param("external_id", ToolParamType.STRING, "外部幂等 ID", True),
+            _tool_param("source_type", ToolParamType.STRING, "来源类型", True),
+            _tool_param("text", ToolParamType.STRING, "原始文本", True),
+            _tool_param("chat_id", ToolParamType.STRING, "聊天流 ID", False),
+            _tool_param("timestamp", ToolParamType.FLOAT, "时间戳", False),
+        ],
+    )
+    async def handle_ingest_text(
+        self,
+        external_id: str,
+        source_type: str,
+        text: str,
+        chat_id: str = "",
+        person_ids: Optional[List[str]] = None,
+        participants: Optional[List[str]] = None,
+        timestamp: float | None = None,
+        time_start: float | None = None,
+        time_end: float | None = None,
+        tags: Optional[List[str]] = None,
+        metadata: Optional[Dict[str, Any]] = None,
+        **kwargs,
+    ):
+        relations = kwargs.get("relations")
+        entities = kwargs.get("entities")
+        kernel = await self._get_kernel()
+        return await kernel.ingest_text(
+            external_id=external_id,
+            source_type=source_type,
+            text=text,
+            chat_id=chat_id,
+            person_ids=person_ids,
+            participants=participants,
+            timestamp=timestamp,
+            time_start=time_start,
+            time_end=time_end,
+            tags=tags,
+            metadata=metadata,
+            entities=entities,
+            relations=relations,
+        )
+
+    @Tool(
+        "get_person_profile",
+        description="获取人物画像",
+        parameters=[
+            _tool_param("person_id", ToolParamType.STRING, "人物 ID", True),
+            _tool_param("chat_id", ToolParamType.STRING, "聊天流 ID", False),
+            _tool_param("limit", ToolParamType.INTEGER, "证据条数", False),
+        ],
+    )
+    async def handle_get_person_profile(self, person_id: str, chat_id: str = "", limit: int = 10, **kwargs):
+        _ = kwargs
+        kernel = await self._get_kernel()
+        return await kernel.get_person_profile(person_id=person_id, chat_id=chat_id, limit=limit)
+
+    @Tool(
+        "maintain_memory",
+        description="维护长期记忆关系状态",
+        parameters=[
+            _tool_param("action", ToolParamType.STRING, "reinforce/protect/restore", True),
+            _tool_param("target", ToolParamType.STRING, "目标哈希或查询文本", True),
+            _tool_param("hours", ToolParamType.FLOAT, "保护时长（小时）", False),
+        ],
+    )
+    async def handle_maintain_memory(
+        self,
+        action: str,
+        target: str,
+        hours: float | None = None,
+        reason: str = "",
+        **kwargs,
+    ):
+        _ = kwargs
+        kernel = await self._get_kernel()
+        return await kernel.maintain_memory(action=action, target=target, hours=hours, reason=reason)
+
+    @Tool("memory_stats", description="获取长期记忆统计", parameters=[])
+    async def handle_memory_stats(self, **kwargs):
+        _ = kwargs
+        kernel = await self._get_kernel()
+        return kernel.memory_stats()
+
+
+def create_plugin():
+    return AMemorixPlugin()

plugins/A_memorix/scripts/convert_lpmm.py

@@ -0,0 +1,535 @@
+#!/usr/bin/env python3
+"""
+LPMM 到 A_memorix 存储转换器
+
+功能：
+1. 读取 LPMM parquet 文件 (paragraph.parquet, entity.parquet, relation.parquet)
+2. 读取 LPMM 图文件 (graph.graphml 或 graph_structure.pkl)
+3. 直接写入 A_memorix 二进制 VectorStore 和稀疏 GraphStore
+4. 绕过 Embedding 生成以节省 Token
+"""
+
+import sys
+import os
+import json
+import argparse
+import asyncio
+import pickle
+import logging
+from pathlib import Path
+from typing import Dict, Any, List, Tuple
+import numpy as np
+import tomlkit
+
+# 设置路径
+current_dir = Path(__file__).resolve().parent
+plugin_root = current_dir.parent
+project_root = plugin_root.parent.parent
+sys.path.insert(0, str(project_root))
+
+def _build_arg_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser(description="将 LPMM 数据转换为 A_memorix 格式")
+    parser.add_argument("--input", "-i", required=True, help="包含 LPMM 数据的输入目录 (parquet, graphml)")
+    parser.add_argument("--output", "-o", required=True, help="A_memorix 数据的输出目录")
+    parser.add_argument("--dim", type=int, default=384, help="Embedding 维度 (必须与 LPMM 模型匹配)")
+    parser.add_argument("--batch-size", type=int, default=1024, help="Parquet 分批读取大小 (默认 1024)")
+    parser.add_argument(
+        "--skip-relation-vector-rebuild",
+        action="store_true",
+        help="跳过按关系元数据重建关系向量（默认开启）",
+    )
+    return parser
+
+
+# --help/-h fast path: avoid heavy host/plugin bootstrap
+if any(arg in {"-h", "--help"} for arg in sys.argv[1:]):
+    _build_arg_parser().print_help()
+    sys.exit(0)
+
+# 设置日志
+logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(name)s - %(levelname)s - %(message)s")
+logger = logging.getLogger("LPMM_Converter")
+
+try:
+    import networkx as nx
+    from scipy import sparse
+    import pyarrow.parquet as pq
+except ImportError as e:
+    logger.error(f"缺少依赖: {e}")
+    logger.error("请安装: pip install pandas pyarrow networkx scipy")
+    sys.exit(1)
+
+try:
+    # 优先采取相对导入 (将插件根目录加入路径)
+    # 这样可以避免硬编码插件名称 (plugins.A_memorix)
+    if str(plugin_root) not in sys.path:
+        sys.path.insert(0, str(plugin_root))
+    
+    from core.storage.vector_store import VectorStore
+    from core.storage.graph_store import GraphStore
+    from core.storage.metadata_store import MetadataStore
+    from core.storage import QuantizationType, SparseMatrixFormat
+    from core.embedding import create_embedding_api_adapter
+    from core.utils.relation_write_service import RelationWriteService
+    
+except ImportError as e:
+    logger.error(f"无法导入 A_memorix 核心模块: {e}")
+    logger.error("请确保在正确的环境中运行，且已安装所有依赖。")
+    sys.exit(1)
+
+
+class LPMMConverter:
+    def __init__(
+        self,
+        lpmm_data_dir: Path,
+        output_dir: Path,
+        dimension: int = 384,
+        batch_size: int = 1024,
+        rebuild_relation_vectors: bool = True,
+    ):
+        self.lpmm_dir = lpmm_data_dir
+        self.output_dir = output_dir
+        self.dimension = dimension
+        self.batch_size = max(1, int(batch_size))
+        self.rebuild_relation_vectors = bool(rebuild_relation_vectors)
+        
+        self.vector_dir = output_dir / "vectors"
+        self.graph_dir = output_dir / "graph"
+        self.metadata_dir = output_dir / "metadata"
+        
+        self.vector_store = None
+        self.graph_store = None
+        self.metadata_store = None
+        self.embedding_manager = None
+        self.relation_write_service = None
+        # LPMM 原 ID -> A_memorix ID 映射（用于图重写）
+        self.id_mapping: Dict[str, str] = {}
+
+    def _register_id_mapping(self, raw_id: Any, mapped_id: str, p_type: str) -> None:
+        """记录 ID 映射，兼容带/不带类型前缀两种格式。"""
+        if raw_id is None:
+            return
+
+        raw = str(raw_id).strip()
+        if not raw:
+            return
+
+        self.id_mapping[raw] = mapped_id
+
+        prefix = f"{p_type}-"
+        if raw.startswith(prefix):
+            self.id_mapping[raw[len(prefix):]] = mapped_id
+        else:
+            self.id_mapping[prefix + raw] = mapped_id
+
+    def _map_node_id(self, node: Any) -> str:
+        """将图节点 ID 映射到转换后的 A_memorix ID。"""
+        node_key = str(node)
+        return self.id_mapping.get(node_key, node_key)
+        
+    def initialize_stores(self):
+        """初始化空的 A_memorix 存储"""
+        logger.info(f"正在初始化存储于 {self.output_dir}...")
+        
+        # 初始化 VectorStore (A_memorix 默认使用 INT8 量化)
+        self.vector_store = VectorStore(
+            dimension=self.dimension,
+            quantization_type=QuantizationType.INT8,
+            data_dir=self.vector_dir
+        )
+        self.vector_store.clear() # 清空旧数据
+        
+        # 初始化 GraphStore (使用 CSR 格式)
+        self.graph_store = GraphStore(
+            matrix_format=SparseMatrixFormat.CSR,
+            data_dir=self.graph_dir
+        )
+        self.graph_store.clear()
+        
+        # 初始化 MetadataStore
+        self.metadata_store = MetadataStore(data_dir=self.metadata_dir)
+        self.metadata_store.connect()
+        # 清空元数据表？理想情况下是的，但要小心。
+        # 对于转换，我们假设是全新的开始或覆盖。
+        # A_memorix 中的 MetadataStore 通常使用 SQLite。
+        # 如果目录是新的，我们会依赖它创建新文件。
+        if self.rebuild_relation_vectors:
+            self._init_relation_vector_service()
+
+    def _load_plugin_config(self) -> Dict[str, Any]:
+        config_path = plugin_root / "config.toml"
+        if not config_path.exists():
+            return {}
+        try:
+            with open(config_path, "r", encoding="utf-8") as f:
+                parsed = tomlkit.load(f)
+            return dict(parsed) if isinstance(parsed, dict) else {}
+        except Exception as e:
+            logger.warning(f"读取 config.toml 失败，使用默认 embedding 配置: {e}")
+            return {}
+
+    def _init_relation_vector_service(self) -> None:
+        if not self.rebuild_relation_vectors:
+            return
+        cfg = self._load_plugin_config()
+        emb_cfg = cfg.get("embedding", {}) if isinstance(cfg, dict) else {}
+        if not isinstance(emb_cfg, dict):
+            emb_cfg = {}
+        try:
+            self.embedding_manager = create_embedding_api_adapter(
+                batch_size=int(emb_cfg.get("batch_size", 32)),
+                max_concurrent=int(emb_cfg.get("max_concurrent", 5)),
+                default_dimension=int(emb_cfg.get("dimension", self.dimension)),
+                model_name=str(emb_cfg.get("model_name", "auto")),
+                retry_config=emb_cfg.get("retry", {}) if isinstance(emb_cfg.get("retry", {}), dict) else {},
+            )
+            self.relation_write_service = RelationWriteService(
+                metadata_store=self.metadata_store,
+                graph_store=self.graph_store,
+                vector_store=self.vector_store,
+                embedding_manager=self.embedding_manager,
+            )
+        except Exception as e:
+            self.embedding_manager = None
+            self.relation_write_service = None
+            logger.warning(f"初始化关系向量重建服务失败，将跳过关系向量回填: {e}")
+
+    async def _rebuild_relation_vectors(self) -> None:
+        if not self.rebuild_relation_vectors:
+            return
+        if self.relation_write_service is None:
+            logger.warning("关系向量重建已启用，但写入服务不可用，已跳过。")
+            return
+
+        rows = self.metadata_store.get_relations()
+        if not rows:
+            logger.info("未发现关系元数据，无需重建关系向量。")
+            return
+
+        success = 0
+        failed = 0
+        skipped = 0
+        for row in rows:
+            result = await self.relation_write_service.ensure_relation_vector(
+                hash_value=str(row["hash"]),
+                subject=str(row.get("subject", "")),
+                predicate=str(row.get("predicate", "")),
+                obj=str(row.get("object", "")),
+            )
+            if result.vector_state == "ready":
+                if result.vector_written:
+                    success += 1
+                else:
+                    skipped += 1
+            else:
+                failed += 1
+
+        logger.info(
+            "关系向量重建完成: total=%s success=%s skipped=%s failed=%s",
+            len(rows),
+            success,
+            skipped,
+            failed,
+        )
+
+    @staticmethod
+    def _parse_relation_text(text: str) -> Tuple[str, str, str]:
+        raw = str(text or "").strip()
+        if not raw:
+            return "", "", ""
+        if "|" in raw:
+            parts = [p.strip() for p in raw.split("|") if p.strip()]
+            if len(parts) >= 3:
+                return parts[0], parts[1], parts[2]
+        if "->" in raw:
+            parts = [p.strip() for p in raw.split("->") if p.strip()]
+            if len(parts) >= 3:
+                return parts[0], parts[1], parts[2]
+        pieces = raw.split()
+        if len(pieces) >= 3:
+            return pieces[0], pieces[1], " ".join(pieces[2:])
+        return "", "", ""
+
+    def _import_relation_metadata_from_parquet(self, relation_path: Path) -> int:
+        if not relation_path.exists():
+            return 0
+
+        try:
+            parquet_file = pq.ParquetFile(relation_path)
+        except Exception as e:
+            logger.warning(f"读取 relation.parquet 失败，跳过关系元数据导入: {e}")
+            return 0
+
+        cols = set(parquet_file.schema_arrow.names)
+        has_triple_cols = {"subject", "predicate", "object"}.issubset(cols)
+        content_col = "str" if "str" in cols else ("content" if "content" in cols else "")
+
+        imported_hashes = set()
+        imported = 0
+        for record_batch in parquet_file.iter_batches(batch_size=self.batch_size):
+            df_batch = record_batch.to_pandas()
+            for _, row in df_batch.iterrows():
+                subject = ""
+                predicate = ""
+                obj = ""
+                if has_triple_cols:
+                    subject = str(row.get("subject", "") or "").strip()
+                    predicate = str(row.get("predicate", "") or "").strip()
+                    obj = str(row.get("object", "") or "").strip()
+                elif content_col:
+                    subject, predicate, obj = self._parse_relation_text(row.get(content_col, ""))
+
+                if not (subject and predicate and obj):
+                    continue
+
+                rel_hash = self.metadata_store.add_relation(
+                    subject=subject,
+                    predicate=predicate,
+                    obj=obj,
+                    source_paragraph=None,
+                )
+                if rel_hash in imported_hashes:
+                    continue
+                imported_hashes.add(rel_hash)
+                self.graph_store.add_edges([(subject, obj)], relation_hashes=[rel_hash])
+                try:
+                    self.metadata_store.set_relation_vector_state(rel_hash, "none")
+                except Exception:
+                    pass
+                imported += 1
+
+        return imported
+        
+    def convert_vectors(self):
+        """将 Parquet 向量转换为 VectorStore"""
+        # LPMM 默认文件名
+        parquet_files = {
+            "paragraph": self.lpmm_dir / "paragraph.parquet",
+            "entity": self.lpmm_dir / "entity.parquet",
+            "relation": self.lpmm_dir / "relation.parquet"
+        }
+        
+        total_vectors = 0
+        
+        for p_type, p_path in parquet_files.items():
+            # 关系向量在当前脚本中无法保证与 MetadataStore 的关系记录一一对应，
+            # 直接导入会污染召回结果（命中后无法反查 relation 元数据）。
+            if p_type == "relation":
+                relation_count = self._import_relation_metadata_from_parquet(p_path)
+                logger.warning(
+                    "跳过 relation.parquet 向量导入（保持一致性）；已导入关系元数据: %s",
+                    relation_count,
+                )
+                continue
+
+            if not p_path.exists():
+                logger.warning(f"文件未找到: {p_path}, 跳过 {p_type} 向量。")
+                continue
+                
+            logger.info(f"正在处理 {p_type} 向量，来源: {p_path}...")
+            try:
+                parquet_file = pq.ParquetFile(p_path)
+                total_rows = parquet_file.metadata.num_rows
+                if total_rows == 0:
+                    logger.info(f"{p_path} 为空，跳过。")
+                    continue
+
+                # LPMM Schema: 'hash', 'embedding', 'str'
+                cols = parquet_file.schema_arrow.names
+                # 兼容性检查
+                content_col = 'str' if 'str' in cols else 'content'
+                emb_col = 'embedding'
+                hash_col = 'hash'
+                
+                if content_col not in cols or emb_col not in cols:
+                    logger.error(f"{p_path} 中缺少必要列 (需包含 {content_col}, {emb_col})。发现: {cols}")
+                    continue
+                
+                batch_columns = [content_col, emb_col]
+                if hash_col in cols:
+                    batch_columns.append(hash_col)
+
+                processed_rows = 0
+                added_for_type = 0
+                batch_idx = 0
+
+                for record_batch in parquet_file.iter_batches(
+                    batch_size=self.batch_size,
+                    columns=batch_columns,
+                ):
+                    batch_idx += 1
+                    df_batch = record_batch.to_pandas()
+
+                    embeddings_list = []
+                    ids_list = []
+
+                    # 同时处理元数据映射
+                    for _, row in df_batch.iterrows():
+                        processed_rows += 1
+                        content = row[content_col]
+                        emb = row[emb_col]
+
+                        if content is None or (isinstance(content, float) and np.isnan(content)):
+                            continue
+                        content = str(content).strip()
+                        if not content:
+                            continue
+
+                        if emb is None or len(emb) == 0:
+                            continue
+
+                        # 先写 MetadataStore，并使用其返回的真实 hash 作为向量 ID
+                        # 保证检索返回 ID 可以直接反查元数据。
+                        store_id = None
+                        if p_type == "paragraph":
+                            store_id = self.metadata_store.add_paragraph(
+                                content=content,
+                                source="lpmm_import",
+                                knowledge_type="factual",
+                            )
+                        elif p_type == "entity":
+                            store_id = self.metadata_store.add_entity(name=content)
+                        else:
+                            continue
+
+                        raw_hash = row[hash_col] if hash_col in df_batch.columns else None
+                        if raw_hash is not None and not (isinstance(raw_hash, float) and np.isnan(raw_hash)):
+                            self._register_id_mapping(raw_hash, store_id, p_type)
+                        
+                        # 确保 embedding 是 numpy 数组
+                        emb_np = np.array(emb, dtype=np.float32)
+                        if emb_np.shape[0] != self.dimension:
+                            logger.error(f"维度不匹配: {emb_np.shape[0]} vs {self.dimension}")
+                            continue
+                            
+                        embeddings_list.append(emb_np)
+                        ids_list.append(store_id)
+                    
+                    if embeddings_list:
+                        # 分批添加到向量存储
+                        vectors_np = np.stack(embeddings_list)
+                        count = self.vector_store.add(vectors_np, ids_list)
+                        added_for_type += count
+                        total_vectors += count
+
+                    if batch_idx == 1 or batch_idx % 10 == 0:
+                        logger.info(
+                            f"[{p_type}] 批次 {batch_idx}: 已扫描 {processed_rows}/{total_rows}, 已导入 {added_for_type}"
+                        )
+
+                logger.info(
+                    f"{p_type} 向量处理完成：总扫描 {processed_rows}，总导入 {added_for_type}"
+                )
+                    
+            except Exception as e:
+                logger.error(f"处理 {p_path} 时出错: {e}")
+                
+        # 提交向量存储
+        self.vector_store.save()
+        logger.info(f"向量转换完成。总向量数: {total_vectors}")
+
+    def convert_graph(self):
+        """将 LPMM 图转换为 GraphStore"""
+        # LPMM 默认文件名是 rag-graph.graphml
+        graph_files = [
+            self.lpmm_dir / "rag-graph.graphml",
+            self.lpmm_dir / "graph.graphml",
+            self.lpmm_dir / "graph_structure.pkl"
+        ]
+        
+        nx_graph = None
+        
+        for g_path in graph_files:
+            if g_path.exists():
+                logger.info(f"发现图文件: {g_path}")
+                try:
+                    if g_path.suffix == ".graphml":
+                        nx_graph = nx.read_graphml(g_path)
+                    elif g_path.suffix == ".pkl":
+                        with open(g_path, "rb") as f:
+                            data = pickle.load(f)
+                            # LPMM 可能会将图存储在包装类中
+                            if hasattr(data, "graph") and isinstance(data.graph, nx.Graph):
+                                nx_graph = data.graph
+                            elif isinstance(data, nx.Graph):
+                                nx_graph = data
+                    break
+                except Exception as e:
+                    logger.error(f"加载 {g_path} 失败: {e}")
+        
+        if nx_graph is None:
+            logger.warning("未找到有效的图文件。跳过图转换。")
+            return
+
+        logger.info(f"已加载图，包含 {nx_graph.number_of_nodes()} 个节点和 {nx_graph.number_of_edges()} 条边。")
+        
+        # 1. 添加节点
+        # LPMM 节点通常是哈希或带前缀的字符串。
+        # 我们需要将它们映射到 A_memorix 格式。
+        # 如果 LPMM 使用 "entity-HASH"，则与 A_memorix 匹配。
+        
+        nodes_to_add = []
+        node_attrs = {}
+        
+        for node, attrs in nx_graph.nodes(data=True):
+            # 假设 LPMM 使用一致的命名 "entity-..." 或 "paragraph-..."
+            mapped_node = self._map_node_id(node)
+            nodes_to_add.append(mapped_node)
+            if attrs:
+                node_attrs[mapped_node] = attrs
+        
+        self.graph_store.add_nodes(nodes_to_add, node_attrs)
+        
+        # 2. 添加边
+        edges_to_add = []
+        weights = []
+        
+        for u, v, data in nx_graph.edges(data=True):
+            weight = data.get("weight", 1.0)
+            edges_to_add.append((self._map_node_id(u), self._map_node_id(v)))
+            weights.append(float(weight))
+            
+            # 如果可能，将关系同步到 MetadataStore
+            # 但图的边并不总是包含关系谓词
+            # 如果 LPMM 边数据有 'predicate'，我们可以添加到元数据
+            # 通常 LPMM 边是加权和，谓词信息可能在简单图中丢失
+            
+        if edges_to_add:
+            self.graph_store.add_edges(edges_to_add, weights)
+            
+        self.graph_store.save()
+        logger.info("图转换完成。")
+
+    def run(self):
+        self.initialize_stores()
+        self.convert_vectors()
+        self.convert_graph()
+        asyncio.run(self._rebuild_relation_vectors())
+        self.vector_store.save()
+        self.graph_store.save()
+        self.metadata_store.close()
+        logger.info("所有转换成功完成。")
+
+
+def main():
+    parser = _build_arg_parser()
+    args = parser.parse_args()
+    
+    input_path = Path(args.input)
+    output_path = Path(args.output)
+    
+    if not input_path.exists():
+        logger.error(f"输入目录不存在: {input_path}")
+        sys.exit(1)
+        
+    converter = LPMMConverter(
+        input_path,
+        output_path,
+        dimension=args.dim,
+        batch_size=args.batch_size,
+        rebuild_relation_vectors=not bool(args.skip_relation_vector_rebuild),
+    )
+    converter.run()
+
+if __name__ == "__main__":
+    main()

plugins/A_memorix/scripts/migrate_chat_history.py

@@ -0,0 +1,110 @@
+#!/usr/bin/env python3
+from __future__ import annotations
+
+import argparse
+import asyncio
+import json
+import sqlite3
+import sys
+from datetime import datetime
+from pathlib import Path
+from typing import Any, Dict
+
+
+CURRENT_DIR = Path(__file__).resolve().parent
+PLUGIN_ROOT = CURRENT_DIR.parent
+WORKSPACE_ROOT = PLUGIN_ROOT.parent
+MAIBOT_ROOT = WORKSPACE_ROOT / "MaiBot"
+DEFAULT_DB_PATH = MAIBOT_ROOT / "data" / "MaiBot.db"
+
+if str(WORKSPACE_ROOT) not in sys.path:
+    sys.path.insert(0, str(WORKSPACE_ROOT))
+if str(MAIBOT_ROOT) not in sys.path:
+    sys.path.insert(0, str(MAIBOT_ROOT))
+
+from A_memorix.core.runtime.sdk_memory_kernel import SDKMemoryKernel  # noqa: E402
+
+
+def _parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description="迁移 MaiBot chat_history 到 A_Memorix")
+    parser.add_argument("--db-path", default=str(DEFAULT_DB_PATH), help="MaiBot SQLite 路径")
+    parser.add_argument("--data-dir", default="./data", help="A_Memorix 数据目录")
+    parser.add_argument("--limit", type=int, default=0, help="限制迁移条数，0 表示全部")
+    parser.add_argument("--dry-run", action="store_true", help="仅预览，不写入")
+    return parser.parse_args()
+
+
+def _to_timestamp(value: Any) -> float | None:
+    if value is None:
+        return None
+    if isinstance(value, (int, float)):
+        return float(value)
+    text = str(value).strip()
+    if not text:
+        return None
+    try:
+        return datetime.fromisoformat(text).timestamp()
+    except ValueError:
+        return None
+
+
+async def _main() -> int:
+    args = _parse_args()
+    db_path = Path(args.db_path).resolve()
+    if not db_path.exists():
+        print(f"数据库不存在: {db_path}")
+        return 1
+
+    conn = sqlite3.connect(str(db_path))
+    conn.row_factory = sqlite3.Row
+    sql = """
+        SELECT id, session_id, start_timestamp, end_timestamp, participants, theme, keywords, summary
+        FROM chat_history
+        ORDER BY id ASC
+    """
+    if int(args.limit or 0) > 0:
+        sql += " LIMIT ?"
+        rows = conn.execute(sql, (int(args.limit),)).fetchall()
+    else:
+        rows = conn.execute(sql).fetchall()
+    conn.close()
+
+    print(f"chat_history 待处理: {len(rows)}")
+    if args.dry_run:
+        for row in rows[:5]:
+            print(f"- id={row['id']} session={row['session_id']} theme={row['theme']}")
+        return 0
+
+    kernel = SDKMemoryKernel(plugin_root=PLUGIN_ROOT, config={"storage": {"data_dir": args.data_dir}})
+    await kernel.initialize()
+    migrated = 0
+    skipped = 0
+    for row in rows:
+        participants = json.loads(row["participants"]) if row["participants"] else []
+        keywords = json.loads(row["keywords"]) if row["keywords"] else []
+        theme = str(row["theme"] or "").strip()
+        summary = str(row["summary"] or "").strip()
+        text = f"主题：{theme}\n概括：{summary}".strip()
+        result: Dict[str, Any] = await kernel.ingest_summary(
+            external_id=f"chat_history:{row['id']}",
+            chat_id=str(row["session_id"] or ""),
+            text=text,
+            participants=participants,
+            time_start=_to_timestamp(row["start_timestamp"]),
+            time_end=_to_timestamp(row["end_timestamp"]),
+            tags=keywords,
+            metadata={"theme": theme, "source_row_id": int(row["id"])},
+        )
+        if result.get("stored_ids"):
+            migrated += 1
+        else:
+            skipped += 1
+
+    print(f"迁移完成: migrated={migrated} skipped={skipped}")
+    print(json.dumps(kernel.memory_stats(), ensure_ascii=False))
+    kernel.close()
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(asyncio.run(_main()))

plugins/A_memorix/scripts/migrate_person_memory_points.py

@@ -0,0 +1,120 @@
+#!/usr/bin/env python3
+from __future__ import annotations
+
+import argparse
+import asyncio
+import json
+import sqlite3
+import sys
+from pathlib import Path
+from typing import Any, Dict, List
+
+
+CURRENT_DIR = Path(__file__).resolve().parent
+PLUGIN_ROOT = CURRENT_DIR.parent
+WORKSPACE_ROOT = PLUGIN_ROOT.parent
+MAIBOT_ROOT = WORKSPACE_ROOT / "MaiBot"
+DEFAULT_DB_PATH = MAIBOT_ROOT / "data" / "MaiBot.db"
+
+if str(WORKSPACE_ROOT) not in sys.path:
+    sys.path.insert(0, str(WORKSPACE_ROOT))
+if str(MAIBOT_ROOT) not in sys.path:
+    sys.path.insert(0, str(MAIBOT_ROOT))
+
+from A_memorix.core.runtime.sdk_memory_kernel import SDKMemoryKernel  # noqa: E402
+
+
+def _parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description="迁移 MaiBot person_info.memory_points 到 A_Memorix")
+    parser.add_argument("--db-path", default=str(DEFAULT_DB_PATH), help="MaiBot SQLite 路径")
+    parser.add_argument("--data-dir", default="./data", help="A_Memorix 数据目录")
+    parser.add_argument("--limit", type=int, default=0, help="限制迁移人数，0 表示全部")
+    parser.add_argument("--dry-run", action="store_true", help="仅预览，不写入")
+    return parser.parse_args()
+
+
+def _parse_memory_points(raw_value: Any) -> List[Dict[str, Any]]:
+    try:
+        values = json.loads(raw_value) if raw_value else []
+    except Exception:
+        values = []
+    items: List[Dict[str, Any]] = []
+    for index, item in enumerate(values):
+        text = str(item or "").strip()
+        if not text:
+            continue
+        parts = text.split(":")
+        if len(parts) >= 3:
+            category = parts[0].strip()
+            content = ":".join(parts[1:-1]).strip()
+            weight = parts[-1].strip()
+        else:
+            category = "其他"
+            content = text
+            weight = "1.0"
+        if content:
+            items.append({"index": index, "category": category or "其他", "content": content, "weight": weight or "1.0"})
+    return items
+
+
+async def _main() -> int:
+    args = _parse_args()
+    db_path = Path(args.db_path).resolve()
+    if not db_path.exists():
+        print(f"数据库不存在: {db_path}")
+        return 1
+
+    conn = sqlite3.connect(str(db_path))
+    conn.row_factory = sqlite3.Row
+    sql = """
+        SELECT person_id, person_name, user_nickname, memory_points
+        FROM person_info
+        WHERE memory_points IS NOT NULL AND memory_points != ''
+        ORDER BY id ASC
+    """
+    if int(args.limit or 0) > 0:
+        sql += " LIMIT ?"
+        rows = conn.execute(sql, (int(args.limit),)).fetchall()
+    else:
+        rows = conn.execute(sql).fetchall()
+    conn.close()
+
+    preview_total = sum(len(_parse_memory_points(row["memory_points"])) for row in rows)
+    print(f"person_info 待迁移人物: {len(rows)} 记忆点: {preview_total}")
+    if args.dry_run:
+        for row in rows[:5]:
+            print(f"- person_id={row['person_id']} person_name={row['person_name'] or row['user_nickname']}")
+        return 0
+
+    kernel = SDKMemoryKernel(plugin_root=PLUGIN_ROOT, config={"storage": {"data_dir": args.data_dir}})
+    await kernel.initialize()
+    migrated = 0
+    skipped = 0
+    for row in rows:
+        person_id = str(row["person_id"] or "").strip()
+        if not person_id:
+            continue
+        display_name = str(row["person_name"] or row["user_nickname"] or "").strip()
+        for item in _parse_memory_points(row["memory_points"]):
+            result: Dict[str, Any] = await kernel.ingest_text(
+                external_id=f"person_memory:{person_id}:{item['index']}",
+                source_type="person_fact",
+                text=f"[{item['category']}] {item['content']}",
+                person_ids=[person_id],
+                tags=[item["category"]],
+                entities=[person_id, display_name] if display_name else [person_id],
+                metadata={"category": item["category"], "weight": item["weight"], "display_name": display_name},
+            )
+            if result.get("stored_ids"):
+                migrated += 1
+            else:
+                skipped += 1
+
+    print(f"迁移完成: migrated={migrated} skipped={skipped}")
+    print(json.dumps(kernel.memory_stats(), ensure_ascii=False))
+    kernel.close()
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(asyncio.run(_main()))