While you sleep, a local LLM writes code, runs it, discovers what works and what doesn't, then throws the code away and keeps only the insights.
Night Bot is a 24/7 AI development crew running entirely on local LLMs. Zero cloud cost.
You (before bed): "Can vitest resolve tsconfig path aliases in a monorepo?"
Night Bot (overnight): writes code → runs in sandbox → error → fix → retry → success
You (morning): reads report.md with findings, gotchas, and next steps
- Scout — writes code, runs it in Docker sandbox, generates report, destroys code
- Planner — 24h task graph with dependencies, assigns work, replans on results
- Secretary — orchestrates everything, escalates to human only when needed
pip install -e .
nightbot setup
nightbot add "Can we migrate from Jest to Vitest without breaking our test suite?"
nightbot start
nightbot briefing- Python 3.11+
- ollama with a coding model (e.g.
qwen2.5-coder:14b) - Docker
See ROADMAP.md for the phase-by-phase plan.
MIT