apps/ai-agents. It runs a LangGraph agent runtime that powers agentic CRM chat and meeting intelligence features.
Prerequisites
- Repo cloned and dependencies installed (
bun install) - Backend running at
http://localhost:3001 OPENAI_API_KEYavailable
Environment variables
Copy the example file:Database
OpenAI
Backend connection
Tracing (optional)
Start the development server
The first start may be slow — the LangGraph CLI downloads additional dependencies on first run.
LangGraph Studio
When the service is running, open LangGraph Studio at http://localhost:8080 to:- Visualize agent execution graphs
- Inspect state at each step
- Replay and debug agent runs
- Test individual graph nodes
Useful commands
Troubleshooting
Service won't start
Service won't start
- Ensure
OPENAI_API_KEYis set inapps/ai-agents/.envor root.env - Ensure
AGENT_TOKENmatches the value inapps/backend/.env - The first start may be slow (downloading LangGraph CLI dependencies) — wait a minute before retrying
Port already in use
Port already in use
Enable LangSmith tracing
Enable LangSmith tracing
Database checkpoint errors
Database checkpoint errors
The agents service uses PostgreSQL for LangGraph checkpointing. Ensure
DATABASE_URL is set and the zuko_dev database exists: