Version: 0.5.6.dev.260314
✨ feat(agent): 重构 Agent 分层并修复普通聊天助手消息未写入 Redis 的问题 🔧 按职责重构 backend/agent 目录为 route/chat/quicknote 三层结构 🔄 将随口记链路拆分为 graph/nodes/tool/state/prompt，其中 graph 仅负责连线 🏃 新增 quicknote runner（方法引用）来收口节点依赖，提升代码可读性 🔀 将控制码分流逻辑抽离到 agent/route，服务层改为薄封装调用 📚 更新相关 README 与测试引用路径，保持原业务逻辑不变 🐛 修复普通聊天链路遗漏 assistant 写入 Redis 的问题（确保 MySQL 和 Redis 的口径一致）
This commit is contained in:
Losita
34
backend/agent/chat/prompt.go
Normal file
34
backend/agent/chat/prompt.go
Normal file
@@ -0,0 +1,34 @@
|
||||
package chat
|
||||
|
||||
const (
|
||||
// SystemPrompt 全局系统人设：定义 SmartFlow 的基本调性
|
||||
SystemPrompt = `你叫 SmartFlow，是专为重邮（CQUPT）学子打造的智能排程专家。
|
||||
你的回复应当专业、干练，偶尔可以带一点程序员式的冷幽默。
|
||||
重要约束：你无法直接写入数据库。除非系统明确告知“任务已落库成功”，否则禁止使用“已安排/已记录/已帮你记下”等完成态表述。`
|
||||
|
||||
// SmartAssistantPrompt 合并了分诊与对话能力的超级提示词
|
||||
SmartAssistantPrompt = `你叫 SmartFlow，是专为重邮（CQUPT）学子打造的智能排程专家。
|
||||
### 你的双重职责：
|
||||
1. **直接对话**：如果用户是闲聊、查询简单信息或进行通用问答，请直接以专业且幽默的口吻回复。
|
||||
2. **决策路由**：如果用户提出需要“安排日程”、“解决冲突”或涉及“3D Atomic TimeGrid”的操作，请在回复中明确你的计划，并准备调用相应的排程工具。
|
||||
### 核心约束：
|
||||
- 始终保持对“稳扎稳打（Steady）模式”的敬畏，压缩率不得超过 15%。
|
||||
- 针对重邮场景（如：红岩网校、南山教学楼）提供有温度的建议。
|
||||
### 输出格式：
|
||||
- 如果涉及排程工具调用，请先简要说明你的调整思路，再执行动作。`
|
||||
|
||||
// SchedulerPromptTemplate 排程专家 (Scheduler)：核心算法 Agent
|
||||
// 这里注入 3D Grid 和 Steady 模式的约束
|
||||
SchedulerPromptTemplate = `你是一位精通“三维原子时间网格（3D Atomic TimeGrid）”的顶级排程架构师。
|
||||
在处理用户的排程请求时，你必须遵循以下硬性逻辑约束：
|
||||
1. 稳扎稳打（Steady）模式：任务步长（Step）的动态分配必须保守，压缩率严禁超过原始时长的 15%。
|
||||
2. 逻辑空间投影（Logical Space Mapping）：当发生时空重叠时，优先尝试在逻辑向量维度平移，而非直接删除冲突任务。
|
||||
3. 冲突自愈：若发现网格冲突，请主动提出“缩放任务块”或“重新锚定时间点”的自愈方案。
|
||||
|
||||
请以极其严谨的态度处理每一秒钟的分配。`
|
||||
|
||||
// DefaultPromptTemplate 通用助手 (Assistant)：也就是你之前占位的那个
|
||||
DefaultPromptTemplate = `你是一位时间管理大师、日程安排专家兼个人助理。
|
||||
你的目标是协助用户高效安排日程。请确保你的回答简洁明了，直接针对用户的需求进行回复。
|
||||
如果用户提到重邮（CQUPT）相关内容（如：南山、红岩网校、卓越工程师班），请表现出你的亲切感。`
|
||||
)
|
||||
198
backend/agent/chat/stream.go
Normal file
198
backend/agent/chat/stream.go
Normal file
@@ -0,0 +1,198 @@
|
||||
package chat
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"io"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
"github.com/cloudwego/eino-ext/components/model/ark"
|
||||
"github.com/cloudwego/eino/schema"
|
||||
"github.com/google/uuid"
|
||||
arkModel "github.com/volcengine/volcengine-go-sdk/service/arkruntime/model"
|
||||
)
|
||||
|
||||
// StreamResponse 是 OpenAI/DeepSeek 兼容的流式 chunk 结构。
|
||||
type StreamResponse struct {
|
||||
ID string `json:"id"`
|
||||
Object string `json:"object"`
|
||||
Created int64 `json:"created"`
|
||||
Model string `json:"model"`
|
||||
Choices []StreamChoice `json:"choices"`
|
||||
}
|
||||
|
||||
type StreamChoice struct {
|
||||
Index int `json:"index"`
|
||||
Delta StreamDelta `json:"delta"`
|
||||
FinishReason *string `json:"finish_reason"`
|
||||
}
|
||||
|
||||
type StreamDelta struct {
|
||||
Role string `json:"role,omitempty"`
|
||||
Content string `json:"content,omitempty"`
|
||||
ReasoningContent string `json:"reasoning_content,omitempty"`
|
||||
}
|
||||
|
||||
// ToOpenAIStream 将单个 Eino chunk 转为 OpenAI 兼容 JSON。
|
||||
func ToOpenAIStream(chunk *schema.Message, requestID, modelName string, created int64, includeRole bool) (string, error) {
|
||||
delta := StreamDelta{}
|
||||
if includeRole {
|
||||
delta.Role = "assistant"
|
||||
}
|
||||
if chunk != nil {
|
||||
delta.Content = chunk.Content
|
||||
delta.ReasoningContent = chunk.ReasoningContent
|
||||
}
|
||||
|
||||
if delta.Role == "" && delta.Content == "" && delta.ReasoningContent == "" {
|
||||
return "", nil
|
||||
}
|
||||
|
||||
dto := StreamResponse{
|
||||
ID: requestID,
|
||||
Object: "chat.completion.chunk",
|
||||
Created: created,
|
||||
Model: modelName,
|
||||
Choices: []StreamChoice{{
|
||||
Index: 0,
|
||||
Delta: delta,
|
||||
FinishReason: nil,
|
||||
}},
|
||||
}
|
||||
jsonBytes, err := json.Marshal(dto)
|
||||
if err != nil {
|
||||
return "", err
|
||||
}
|
||||
return string(jsonBytes), nil
|
||||
}
|
||||
|
||||
// ToOpenAIFinishStream 生成结束 chunk（finish_reason=stop）。
|
||||
func ToOpenAIFinishStream(requestID, modelName string, created int64) (string, error) {
|
||||
stop := "stop"
|
||||
dto := StreamResponse{
|
||||
ID: requestID,
|
||||
Object: "chat.completion.chunk",
|
||||
Created: created,
|
||||
Model: modelName,
|
||||
Choices: []StreamChoice{{
|
||||
Index: 0,
|
||||
Delta: StreamDelta{},
|
||||
FinishReason: &stop,
|
||||
}},
|
||||
}
|
||||
jsonBytes, err := json.Marshal(dto)
|
||||
if err != nil {
|
||||
return "", err
|
||||
}
|
||||
return string(jsonBytes), nil
|
||||
}
|
||||
|
||||
// StreamChat 负责模型流式输出，并在关键节点打点：
|
||||
// 1) 流连接建立（llm.Stream 返回）
|
||||
// 2) 首包到达（首字延迟）
|
||||
// 3) 流式输出结束
|
||||
func StreamChat(
|
||||
ctx context.Context,
|
||||
llm *ark.ChatModel,
|
||||
modelName string,
|
||||
userInput string,
|
||||
ifThinking bool,
|
||||
chatHistory []*schema.Message,
|
||||
outChan chan<- string,
|
||||
traceID string,
|
||||
chatID string,
|
||||
requestStart time.Time,
|
||||
) (string, error) {
|
||||
/*callStart := time.Now()*/
|
||||
|
||||
messages := make([]*schema.Message, 0)
|
||||
messages = append(messages, schema.SystemMessage(SystemPrompt))
|
||||
if len(chatHistory) > 0 {
|
||||
messages = append(messages, chatHistory...)
|
||||
}
|
||||
messages = append(messages, schema.UserMessage(userInput))
|
||||
|
||||
var thinking *ark.Thinking
|
||||
if ifThinking {
|
||||
thinking = &arkModel.Thinking{Type: arkModel.ThinkingTypeEnabled}
|
||||
} else {
|
||||
thinking = &arkModel.Thinking{Type: arkModel.ThinkingTypeDisabled}
|
||||
}
|
||||
|
||||
/*connectStart := time.Now()*/
|
||||
reader, err := llm.Stream(ctx, messages, ark.WithThinking(thinking))
|
||||
if err != nil {
|
||||
return "", err
|
||||
}
|
||||
defer reader.Close()
|
||||
|
||||
if strings.TrimSpace(modelName) == "" {
|
||||
modelName = "smartflow-worker"
|
||||
}
|
||||
requestID := "chatcmpl-" + uuid.NewString()
|
||||
created := time.Now().Unix()
|
||||
firstChunk := true
|
||||
chunkCount := 0
|
||||
/*streamRecvStart := time.Now()
|
||||
|
||||
log.Printf("打点|流连接建立|trace_id=%s|chat_id=%s|request_id=%s|本步耗时_ms=%d|请求累计_ms=%d|history_len=%d",
|
||||
traceID,
|
||||
chatID,
|
||||
requestID,
|
||||
time.Since(connectStart).Milliseconds(),
|
||||
time.Since(requestStart).Milliseconds(),
|
||||
len(chatHistory),
|
||||
)*/
|
||||
|
||||
var fullText strings.Builder
|
||||
for {
|
||||
chunk, err := reader.Recv()
|
||||
if err == io.EOF {
|
||||
break
|
||||
}
|
||||
if err != nil {
|
||||
return "", err
|
||||
}
|
||||
|
||||
fullText.WriteString(chunk.Content)
|
||||
|
||||
payload, err := ToOpenAIStream(chunk, requestID, modelName, created, firstChunk)
|
||||
if err != nil {
|
||||
return "", err
|
||||
}
|
||||
if payload != "" {
|
||||
outChan <- payload
|
||||
chunkCount++
|
||||
/*if firstChunk {
|
||||
log.Printf("打点|首包到达|trace_id=%s|chat_id=%s|request_id=%s|本步耗时_ms=%d|请求累计_ms=%d",
|
||||
traceID,
|
||||
chatID,
|
||||
requestID,
|
||||
time.Since(streamRecvStart).Milliseconds(),
|
||||
time.Since(requestStart).Milliseconds(),
|
||||
)
|
||||
firstChunk = false
|
||||
}*/
|
||||
}
|
||||
}
|
||||
|
||||
finishChunk, err := ToOpenAIFinishStream(requestID, modelName, created)
|
||||
if err != nil {
|
||||
return "", err
|
||||
}
|
||||
outChan <- finishChunk
|
||||
outChan <- "[DONE]"
|
||||
|
||||
/*log.Printf("打点|流式输出结束|trace_id=%s|chat_id=%s|request_id=%s|chunks=%d|reply_chars=%d|本步耗时_ms=%d|请求累计_ms=%d",
|
||||
traceID,
|
||||
chatID,
|
||||
requestID,
|
||||
chunkCount,
|
||||
len(fullText.String()),
|
||||
time.Since(callStart).Milliseconds(),
|
||||
time.Since(requestStart).Milliseconds(),
|
||||
)*/
|
||||
|
||||
return fullText.String(), nil
|
||||
}
|
||||
Reference in New Issue
Block a user