feat:同步本地非算法改动到上游基线

保留反馈纠错、WebUI 与运行时增强。\n移除不应提交的 algorithm_redesign 设计目录及其专项测试。

src/A_memorix/core/embedding/api_adapter.py

@@ -11,7 +11,7 @@ from __future__ import annotations
 
 import asyncio
 import time
-from typing import Any, List, Optional, Union
+from typing import Any, Dict, List, Optional, Tuple, Union
 
 import aiohttp
 import numpy as np
@@ -29,6 +29,9 @@ logger = get_logger("A_Memorix.EmbeddingAPIAdapter")
 class EmbeddingAPIAdapter:
     """适配宿主 embedding 请求接口。"""
 
+    _GLOBAL_DIMENSION_CACHE: Dict[str, int] = {}
+    _GLOBAL_TEXT_EMBEDDING_CACHE: Dict[Tuple[str, int, str], np.ndarray] = {}
+
     def __init__(
         self,
         batch_size: int = 32,
@@ -232,10 +235,32 @@ class EmbeddingAPIAdapter:
             logger.error(f"通过直接 Client 获取 Embedding 失败: {last_exc}")
         return None
 
+    def _dimension_cache_key(self) -> str:
+        candidate_names = self._resolve_candidate_model_names()
+        return "|".join(
+            [
+                str(self.model_name or "auto"),
+                str(self.default_dimension),
+                ",".join(candidate_names),
+            ]
+        )
+
+    def _embedding_cache_key(self, text: str, dimensions: Optional[int]) -> Tuple[str, int, str]:
+        requested_dimension = self._resolve_canonical_dimension(dimensions)
+        return (self._dimension_cache_key(), int(requested_dimension), str(text or ""))
+
     async def _detect_dimension(self) -> int:
         if self._dimension_detected and self._dimension is not None:
             return self._dimension
 
+        cache_key = self._dimension_cache_key()
+        cached_dimension = self._GLOBAL_DIMENSION_CACHE.get(cache_key)
+        if cached_dimension is not None:
+            self._dimension = int(cached_dimension)
+            self._dimension_detected = True
+            logger.info(f"嵌入维度命中进程缓存: {self._dimension}")
+            return self._dimension
+
         logger.info("正在检测嵌入模型维度...")
         try:
             target_dim = self.default_dimension
@@ -251,6 +276,7 @@ class EmbeddingAPIAdapter:
                     )
                 self._dimension = detected_dim
                 self._dimension_detected = True
+                self._GLOBAL_DIMENSION_CACHE[cache_key] = int(detected_dim)
                 return detected_dim
         except Exception as exc:
             logger.debug(f"带维度参数探测失败: {exc}，尝试不带维度参数探测")
@@ -261,6 +287,7 @@ class EmbeddingAPIAdapter:
                 detected_dim = len(test_embedding)
                 self._dimension = detected_dim
                 self._dimension_detected = True
+                self._GLOBAL_DIMENSION_CACHE[cache_key] = int(detected_dim)
                 logger.info(f"嵌入维度检测成功 (自然维度): {detected_dim}")
                 return detected_dim
             logger.warning(f"嵌入维度检测失败，使用 configured_dimension: {self.default_dimension}")
@@ -269,6 +296,7 @@ class EmbeddingAPIAdapter:
 
         self._dimension = self.default_dimension
         self._dimension_detected = True
+        self._GLOBAL_DIMENSION_CACHE[cache_key] = int(self.default_dimension)
         return self.default_dimension
 
     async def encode(
@@ -336,6 +364,25 @@ class EmbeddingAPIAdapter:
         all_embeddings: List[np.ndarray] = []
         for offset in range(0, len(texts), batch_size):
             batch = texts[offset : offset + batch_size]
+            batch_results: List[Tuple[int, np.ndarray]] = []
+            uncached_items: List[Tuple[int, str]] = []
+
+            if self.enable_cache:
+                for index, text in enumerate(batch):
+                    cache_key = self._embedding_cache_key(text, dimensions)
+                    cached_vector = self._GLOBAL_TEXT_EMBEDDING_CACHE.get(cache_key)
+                    if cached_vector is None:
+                        uncached_items.append((index, text))
+                    else:
+                        batch_results.append((index, cached_vector.copy()))
+            else:
+                uncached_items = list(enumerate(batch))
+
+            if not uncached_items:
+                batch_results.sort(key=lambda item: item[0])
+                all_embeddings.extend(emb for _, emb in batch_results)
+                continue
+
             semaphore = asyncio.Semaphore(self.max_concurrent)
 
             async def encode_with_semaphore(text: str, index: int):
@@ -351,11 +398,20 @@ class EmbeddingAPIAdapter:
 
             tasks = [
                 encode_with_semaphore(text, offset + index)
-                for index, text in enumerate(batch)
+                for index, text in uncached_items
             ]
             results = await asyncio.gather(*tasks)
-            results.sort(key=lambda item: item[0])
-            all_embeddings.extend(emb for _, emb in results)
+            normalized_results: List[Tuple[int, np.ndarray]] = []
+            for batch_index, vector in results:
+                normalized_results.append((batch_index, vector))
+                if self.enable_cache:
+                    text = batch[batch_index]
+                    cache_key = self._embedding_cache_key(text, dimensions)
+                    self._GLOBAL_TEXT_EMBEDDING_CACHE[cache_key] = vector.copy()
+
+            batch_results.extend(normalized_results)
+            batch_results.sort(key=lambda item: item[0])
+            all_embeddings.extend(emb for _, emb in batch_results)
 
         return np.array(all_embeddings, dtype=np.float32)
 

src/A_memorix/core/storage/metadata_store.py

@@ -34,7 +34,7 @@ except Exception:
 logger = get_logger("A_Memorix.MetadataStore")
 
 
-SCHEMA_VERSION = 10
+SCHEMA_VERSION = 12
 RUNTIME_AUTO_MIGRATION_MIN_SCHEMA_VERSION = 9
 
 

src/A_memorix/scripts/release_vnext_migrate.py

@@ -375,6 +375,30 @@ def _preflight_impl(config_path: Path, data_dir: Path) -> Dict[str, Any]:
                             "memory_feedback_tasks rollback columns missing under current schema version",
                         )
                     )
+                elif not has_stale_marks:
+                    checks.append(
+                        CheckItem(
+                            "CP-15",
+                            "error",
+                            "paragraph_stale_relation_marks table missing under current schema version",
+                        )
+                    )
+                elif not has_profile_refresh_queue:
+                    checks.append(
+                        CheckItem(
+                            "CP-16",
+                            "error",
+                            "person_profile_refresh_queue table missing under current schema version",
+                        )
+                    )
+                elif not has_feedback_rollback_status or not has_feedback_rollback_plan:
+                    checks.append(
+                        CheckItem(
+                            "CP-17",
+                            "error",
+                            "memory_feedback_tasks rollback columns missing under current schema version",
+                        )
+                    )
 
             if _sqlite_table_exists(conn, "relations"):
                 row = conn.execute("SELECT COUNT(*) FROM relations").fetchone()

src/config/official_configs.py

@@ -145,23 +145,23 @@ class VisualConfig(ConfigBase):
     __ui_label__ = "视觉"
     __ui_icon__ = "image"
 
-    planner_mode: Literal["text", "multimodal", "auto"] = Field(
-        default="auto",
+    multimodal_planner: bool = Field(
+        default=True,
         json_schema_extra={
-            "x-widget": "select",
-            "x-icon": "git-branch",
+            "x-widget": "switch",
+            "x-icon": "image",
         },
     )
-    """规划器模式，auto根据模型信息自动选择，text为纯文本模式，multimodal为多模态模式"""
+    """是否直接输入图片"""
 
-    replyer_mode: Literal["text", "multimodal", "auto"] = Field(
-        default="auto",
+    multimodal_replyer: bool = Field(
+        default=False,
         json_schema_extra={
-            "x-widget": "select",
+            "x-widget": "switch",
             "x-icon": "git-branch",
         },
     )
-    """回复器模式，auto根据模型信息自动选择，text为纯文本模式，multimodal为多模态模式"""
+    """是否启用 Maisaka 多模态 replyer 生成器"""
 
     visual_style: str = Field(
         default="请用中文描述这张图片的内容。如果有文字，请把文字描述概括出来，请留意其主题，直观感受，输出为一段平文本，最多30字，请注意不要分点，就输出一段文本",
@@ -239,12 +239,17 @@ class ChatConfig(ConfigBase):
     )
     """Planner 连续被新消息打断的最大次数，0 表示不启用打断"""
 
+    plan_reply_log_max_per_chat: int = Field(
+        default=1024,
+        json_schema_extra={
+            "x-widget": "input",
+            "x-icon": "file-text",
+        },
+    )
+    """每个聊天流最大保存的Plan/Reply日志数量，超过此数量时会自动删除最老的日志"""
+
     group_chat_prompt: str = Field(
-        default="""
-你正在qq群里聊天，下面是群里正在聊的内容，其中包含聊天记录和聊天中的图片。
-回复尽量简短一些。最好一次对一个话题进行回复，免得啰嗦或者回复内容太乱。请注意把握聊天内容。
-不要回复的太频繁！控制回复的频率，不要每个人的消息都回复，只回复你感兴趣的或者主动提及你的。
-""",
+        default="你需要控制自己发言的频率，如果是一对一聊天，可以以较均匀的频率发言；如果用户较多，不要每句都回复，控制回复频率，不要回复的太频繁！控制回复的频率，不要每个人的消息都回复。",
         json_schema_extra={
             "x-widget": "textarea",
             "x-icon": "users",
@@ -253,11 +258,7 @@ class ChatConfig(ConfigBase):
     """_wrap_群聊通用注意事项"""
 
     private_chat_prompts: str = Field(
-        default="""
-你正在聊天，下面是正在聊的内容，其中包含聊天记录和聊天中的图片。
-回复尽量简短一些。请注意把握聊天内容。
-请考虑对方的发言频率，想法，思考自己何时回复以及回复内容。
-""",
+        default="你需要控制自己发言的频率，可以以较均匀的频率发言。",
         json_schema_extra={
             "x-widget": "textarea",
             "x-icon": "user",
@@ -663,15 +664,6 @@ class LearningItem(ConfigBase):
     )
     """是否启用jargon学习"""
 
-    advanced_chosen: bool = Field(
-        default=False,
-        json_schema_extra={
-            "x-widget": "switch",
-            "x-icon": "sparkles",
-        },
-    )
-    """是否启用基于子代理的二次表达方式选择"""
-
 
 class ExpressionGroup(ConfigBase):
     """表达互通组配置类，若列表为空代表全局共享"""
@@ -701,7 +693,6 @@ class ExpressionConfig(ConfigBase):
                 use_expression=True,
                 enable_learning=True,
                 enable_jargon_learning=True,
-                advanced_chosen=False,
             )
         ],
         json_schema_extra={
@@ -1573,6 +1564,35 @@ class MaiSakaConfig(ConfigBase):
     )
     """MaiSaka 使用的用户名称"""
 
+    tool_filter_task_name: str = Field(
+        default="utils",
+        json_schema_extra={
+            "x-widget": "input",
+            "x-icon": "sparkles",
+        },
+    )
+    """工具筛选预判使用的模型任务名"""
+
+    tool_filter_threshold: int = Field(
+        default=20,
+        ge=1,
+        json_schema_extra={
+            "x-widget": "input",
+            "x-icon": "filter",
+        },
+    )
+    """当可用工具总数超过该阈值时，先进行一轮工具筛选"""
+
+    tool_filter_max_keep: int = Field(
+        default=5,
+        ge=1,
+        json_schema_extra={
+            "x-widget": "input",
+            "x-icon": "list-filter",
+        },
+    )
+    """工具筛选阶段最多保留的非内置工具数量"""
+
     show_image_path: bool = Field(
         default=True,
         json_schema_extra={