Backend

AI-powered triage service for open-source GitHub issues. It ingests issues, runs them through an agent pipeline, and drafts responses grounded in a retrieval corpus.

Stack

• FastAPI — HTTP API
• Celery + Redis — background workers and scheduled polling
• PostgreSQL + pgvector — data store and vector search
• SQLAlchemy + Alembic — ORM and migrations
• LangChain / LangGraph — agent orchestration
• Anthropic / OpenAI / Voyage — LLMs and embeddings

Layout

Path	Purpose
api/	FastAPI app, routes, CRUD, and services
agent/	Agent pipeline: intake → gate → research → draft → escalate
rag/	Chunking, embeddings, and retrieval
ingestion/	GitHub issue ingestion
workers/	Celery app and tasks (issue polling, corpus backfill)
database/	Engine, models, and session management
migrations/	Alembic migration scripts
mcp/	MCP client and server

Getting started

Dependencies are managed with uv.

uv sync

Create a .env file with the required secrets (database URL, Redis URL, GitHub token, and LLM API keys).

Run database migrations:

uv run alembic upgrade head

Start the API:

uv run uvicorn api.server:app --reload

Start the workers:

uv run celery -A workers.celery_app worker --beat

Docker

The full stack (Postgres, Redis, API, workers, frontend) is defined in the root docker-compose.yml:

docker compose up