Skip to content

CLAUDE.md

Latest commit

 

History

History
68 lines (47 loc) · 2.76 KB

CLAUDE.md

File metadata and controls

68 lines (47 loc) · 2.76 KB

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project Overview

A RAG (Retrieval-Augmented Generation) intelligent Q&A system for 飞享IM (FshareChat), built with LangGraph state machine orchestration and Claude API. Answers questions about FshareChat features, deployment, and usage via both CLI and HTTP interfaces.

Setup

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env  # then add ANTHROPIC_API_KEY

Running

# Interactive CLI (builds vector store on first run, takes 1-2 min)
python cli.py

# HTTP service (http://localhost:8000, auto-generates /docs)
python service.py

# Rebuild vector store from scraped + static data
python ingest.py

Architecture

The system follows a multi-stage RAG pipeline orchestrated by a LangGraph state machine in graph.py:

Question → [classify] → on_topic → [retrieve] → [grade_docs] → [generate] (streaming)
                      → off_topic → [reject]
                                            → no relevant docs → [fallback]

State (QAState): question, retrieved_docs, relevant_docs, answer, route

Key design choices:

  • Dual knowledge sources: ingest.py combines live web scraping (4 FshareChat URLs) with the large static knowledge block in config.py. The static block is the primary fallback.
  • Per-document grading: grade_docs node calls Claude individually for each retrieved chunk — quality filtering over raw retrieval.
  • Adaptive thinking: All Claude calls use claude-opus-4-7 with thinking={"type": "adaptive"}.
  • Local embeddings: BAAI/bge-small-zh-v1.5 runs locally for Chinese-optimized embeddings; ChromaDB persists to ./chroma_db/.
  • Retriever binding via closure: build_graph(retriever) closes over the retriever so node functions stay pure.

Entry points:

  • cli.py — loads vectorstore, builds graph, calls graph.invoke()
  • service.py — FastAPI with POST /ask (sync), POST /stream (SSE), GET /health; vectorstore initialized in lifespan

Configuration (config.py): API key, model name, ChromaDB path, embedding model name, scrape URLs, and the static knowledge base string.

HTTP API

curl -X POST http://localhost:8000/ask -H "Content-Type: application/json" -d '{"question": "飞享IM支持哪些平台？"}'
# Response: {"question": "...", "answer": "...", "route": "on_topic|off_topic"}

curl -X POST http://localhost:8000/stream -H "Content-Type: application/json" -d '{"question": "..."}'
# SSE stream, terminated by "data: [DONE]"

No Test Suite

There is no automated test suite. The project uses manual curl testing. No linting or CI/CD configuration exists.