Skip to main content

AI Agents

The apps/graphs service is a Python LangGraph microservice for AI agent orchestration.

Architecture​

API keys (OpenAI, Tavily) are configured by the user in the client settings UI. The client sends requests to the NestJS API, which proxies them to the LangGraph service — injecting the keys at the server level.

Client (settings UI)
  │  API keys stored per user
  ▼
NestJS API ──HTTP + keys──▶ Python LangGraph Service
                                  │
                            ┌─────▼──────┐
                            │ StateGraph │
                            │  ┌──────┐  │
                            │  │search│  │  ← Tavily
                            │  └──┬───┘  │
                            │  ┌──▼───┐  │
                            │  │ LLM  │  │  ← OpenAI
                            │  └──────┘  │
                            └────────────┘

Setup​

1. Install dependencies

cd apps/graphs
pip install -r requirements.txt

2. Build the image

langgraph build -t my-saas-graphs:latest

3. Run

pnpm graphs:dev

Key Files​

  • apps/graphs/scraper_graph.py — Example graph (web search + LLM analysis)
  • apps/graphs/langgraph.json — LangGraph config

Adding a New Graph​

  1. Create a new Python file in apps/graphs/
  2. Define a StateGraph with typed state (TypedDict)
  3. Add nodes and edges
  4. Compile and expose via FastAPI or LangGraph server