ARCHON is a submission to the Nebius x NVIDIA Global AI Hackathon. Contributions are welcome as long as they respect the hackathon rules: all code must be original or properly licensed, and the project must keep running on Nebius Token Factory with NVIDIA open models.
0.docs/ Research and design documents
build/ PRD, DESIGN, AGENTS, PLAN, TESTING, SESSIONS
1.platform/
client/ Next.js 16 cockpit
server/ FastAPI orchestrator
2.submission/ Devpost text and the developer feedback log
mainis what gets submitted. Work on branches namedfeat/...,fix/...,docs/..., orchore/....- Conventional Commits:
feat(server): add sandbox fork per candidate,fix(client): reconnect SSE on drop,docs(plan): move golden set to week 5. - Open a pull request with what changed, why, any new environment variables, and how you verified it.
- Update 0.docs/build/SESSIONS.md at the end of every working session.
- When you hit friction with Nebius, Sandboxes, or Nemotron, add a row to 2.submission/FEEDBACK.md the same day.
Never commit .env files or keys. Copy .env.example to .env and fill it locally. The keys in use are NEBIUS_API_KEY, TAVILY_API_KEY, and ARCHON_DEMO_TOKEN.
cd 1.platform/server && ruff check . && ruff format --check . && mypy --strict app/ && pytest tests/unit -q
cd 1.platform/client && npm run lint && npm run typecheckIntegration tests need real keys and are skipped without them. See 0.docs/build/TESTING.md.
The plan has a cut list. If a change adds a dependency, a service, or a mission type that is not in 0.docs/build/PLAN.md, raise it in the PR description first.