Files
group_xinghuo_jinrong/docs/memory/FLOW.md
T
zhanghongyu_0626 1ddd44a6cb Add core database support and enhance documentation
- Introduced `MYSQL_CORE_DATABASE` in `.env.example` and `settings.py` for core database configuration.
- Added `CoreReadOnlyRepository` for read-only access to the `jinrong_core` database.
- Updated `AGENTS.md`, `README.md`, and various documentation files to reflect new agent onboarding processes and project structure.
- Revised requirements in `requirements.txt` to include `langgraph` and `langchain-core`.
- Enhanced `FLOW.md` with local bootstrap instructions for setting up the core simulation environment.
- Added new scripts for database creation and seeding for the core simulation library.
- Improved overall documentation for clarity on project architecture and memory management.
- Updated `TODO.md` to reflect current development priorities and tasks.
2026-09-05 17:39:16 +08:00

120 lines
3.8 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# 实现流程
> To-Be 端到端链路;As-Is 见 `ENVIRONMENT.md`
> 画像分层见 `docs/项目框架设计/表设计/00-架构总览.md`
------
## 0. 本地 bootstrap(新环境 / 新 Agent)
```text
前置:MySQL 8、Redis、Neo4j Desktop、Ollama(bge-m3) 已安装并启动
① 配置
cp .env.example .env
必填:MYSQL_PASSWORD、NEO4J_PASSWORD、DEEPSEEK_API_KEY(对话阶段)
② 依赖
pip install -r requirements.txt
③ Core 模拟库(L0)
.\scripts\core\reset.ps1
→ 创建 jinrong_core + 28 客户 / 12 产品 / 持仓交易种子
④ Agent 共用底座(若库未建)
mysql -u root -p < docs/项目框架设计/表设计/01-mysql-共用底座.sql
⑤ 同步
python scripts/sync/sync_advisor_rel.py → customer_advisor_rel
python scripts/sync/sync_neo4j.py → 关系图
⑥ 验证
uvicorn app.main:app --reload
GET http://127.0.0.1:8000/health → {"status":"ok"}
(可选)Python 调 CoreReadOnlyRepository.get_customer_l0("CUST-10001")
RBAC 联调账号:scripts/dev/rbac-seed-reference.md
```
**尚未自动化:** agent 专用表 SQL、Milvus 建 Collection、JWT 中间件 — 见 `TODO.md`。
------
## 1. 链路总览(目标)
```text
Client → Gateway(JWT/RBAC) → api/chat → agent_service(LangGraph) → Tools → 存储 → 响应 + audit_log
```
当前:`main.py` 仅 health;全链路待 Wave 0 起逐步挂载。
------
## 2. 主链路 · 对话
```text
输入:Authorization + X-Agent-Type + X-Trace-Id + 用户消息
↓
Gateway:验签、角色准入、注入 AuthContext 【待做 T-01】
↓
api/chat:SessionGuard;创建/续 agent_session 【待做】
↓
memory_service:Redis 读最近 N 轮;异步写 agent_message 【待做】
↓
agent_service:LangGraph StateGraph + DeepSeek;按 Agent 类型选 Tool 节点 【待做】
↓
Tool 示例:
- CoreReadOnlyRepository:持仓/流水/L0(自动注入 customer_id) 【Repository 已有,Tool 未接】
- milvus_tool:产品规则 RAG + source_refs 【待做】
- memory_service:读/写 L1/L2/L3(ProfileGuard) 【待做】
- 风控 service 账号:R-02 适当性(无会话) 【待做】
↓
输出:assistant 消息 + has_disclaimer(客户/对外)
↓
落库:agent_message、agent_tool_call、audit_log(同 trace_id)
```
------
## 3. 主链路 · 知识库入库
```text
admin/knowledge 上传 → document_parser → embedding_tool(Ollama bge-m3)
→ milvus_tool 写入 Collection(dim=1024)
→ 元数据可选写 MySQL;原文件存 data/kb/
```
------
## 4. 分支 · 四 Agent
| X-Agent-Type | 写画像 | 特殊 |
| --- | --- | --- |
| customer | L1 | 仅本人;读 suitability 本人 |
| advisor | L2 | customer_advisor_rel 校验 |
| analyst | 无 | 仅 SELECT SQL |
| risk | L3、预警、适当性 | R-02 可 blocked |
------
## 5. 每步 I/O
| 步骤 | 输入 | 输出 | 约定 |
| --- | --- | --- | --- |
| 鉴权 | JWT | AuthContext | 失败 401/403 + audit |
| 会话 | session_id | Redis ctx + MySQL session | actor_id 一致 |
| RAG | question + product_id? | chunks + source_doc_id | 必须溯源 |
| 画像读 | customer_id | L1/L2/L3 JSON | RBAC 矩阵 §8 |
| Core RO | customer_id | L0/持仓/流水 | 只读 jinrong_core |
| 审计 | 任意判定 | audit_log INSERT | 不可删改 |
------
## 6. 关键数据约定
- Agent 库 SQL:`docs/项目框架设计/表设计/01-mysql-共用底座.sql`、`02-mysql-agent专用.sql`
- Core 模拟:`scripts/core/01-ddl.sql` · 说明 `docs/项目框架设计/Core模拟底座/01-表结构与种子说明.md`
- Redis Key:`docs/项目框架设计/表设计/02-redis-keys.md`
- Milvus:`kb_product_rules`、`kb_business_ops`;向量 **1024** 维
- 禁止:Agent 覆盖 L0 正式 C1~C5