How iamtomshaw/project-manager-agent works
Project type and purpose
• This is an AI-driven, agent-based task and meeting management MVP.
• It’s built as a small microservices/full-stack prototype:
– A FastAPI main service exposing REST and WebSocket endpoints (Chat Agent + webhook)
– A separate Task Manager Agent service (WebSocket microservice)
– A simple Streamlit frontend (frontend.py) for manual testing
– A MongoDB data store and optional Notion integration
File structure
Top-level files and dirs:
• main.py – FastAPI app entrypoint (mounts routes/chat.py, routes/meetings-webhook.py)
• routes/ – WebSocket and webhook endpoints
– chat.py: /ws/chat endpoint implementing the Chat Agent
– meetings-webhook.py: /webhook endpoint for meeting notifications
• lib/ – core libraries
– agent.py: orchestrates calls to tools, maintains agent logic
– ws_manager.py & task_manager.py: WebSocket client/server integrations
– tools.py: function tools exposed to the Chat Agent
– mongo.py: MongoDB helpers
– external.py: Notion or other external-service integrations
• frontend.py – Streamlit UI
• requirements.txt, .env.example – dependencies and environment config
README for additional insights
• Two cooperating agents: Chat Agent (interactive, WS) and Meeting→Project Agent (specialist, WS)
• Chat Agent calls tools in lib/tools.py for project/task operations or to forward messages to the Task Manager Agent
• Meeting webhook triggers the Meeting→Project Agent to assign transcripts to projects
• MongoDB is the primary data store, Notion is optional
Detail level
• Endpoint, tool call, and data store interaction
• Don’t over-aggregate: show lib/tools.py as the “Function Tools” component, lib/mongo.py as “MongoDB Adapter,” etc.
• Related files under their component for clarity