A slice-of-life artifact from the SuperInstance era: a modular AI orchestration platform (AutoGen + FastAPI + Celery + React) for running, watching, and shipping multi-agent writer teams.
What this is: A self-contained snapshot of a working hybrid AI lab we actually ran. Planner → Executor → Refinement workers, a runs API with SSE streaming, an autogen PR flow, a budget-aware model router, and a React frontend to watch it all happen.
What it isn't: Maintained. We're not running this anymore. The shapes are
still useful — see STATUS.md for the salvageable primitives and the
historical context.
make install # Python venv + deps + pre-commit
make run-backend # FastAPI on :8000
make run-worker # Celery worker (separate terminal)
make run-frontend # React UI on :3000 (separate terminal)A compose file at backend/orchestrator/compose/docker-compose.yml brings up
Postgres + Redis + the API if you want the whole stack in one shot.
This repo does not bundle model weights. Set OPENAI_API_KEY /
ANTHROPIC_API_KEY in your .env (see .env.example). The agents call
out to hosted models; local-mode needs a separate model server.
STATUS.md— slice-of-life notes, what was good, what we kept, what's worth stealing.README.original.md— the original project README, preserved verbatim.docs/— design notes from the time (API.md,AUDIT.md,IMPROVEMENTS.md,INDEX.md).
MIT. See LICENSE.
SuperInstance/hybrid-lab— this repoSuperInstance/multiagent-project-archive— a sibling agent-system we archived the same weekSuperInstance/circuit-breaker-archive— separate fault-tolerance lib the team was exploring alongside this
Take this, fork it, evolve it. No attribution required. We'd love to hear what you build, but you owe us nothing.