intelixlab / tools / nightbrain

NIGHTBRAIN(1)

NAME

nightbrain — nightly briefing + one lesson, written by your second brain, delivered before you wake.

SYNOPSIS

git clone https://github.com/nhantour/nightbrain && cd nightbrain
mkdir -p ~/.nightbrain && cp config.env.example ~/.nightbrain/config.env
./install.sh && ./nightbrain.py

DESCRIPTION

every night, on your own hardware:

• runs your signal collectors — any executables printing JSON (failed services, yesterday's commits, your leads file, your sales snapshot)
• writes a morning briefing grounded ONLY in those signals + your vault's GLOBAL.md (north star, rhythm, open loops)
• teaches one lesson from your own notes — rotating sources, skipping anything taught in the last 7 days, ending with a 15-minute drill
• commits both to your vault (git push → your wiki rebuilds) and sends both to your telegram

HOUSE RULE

an empty day is reported as an empty day. the prompt forbids invented wins. a briefing you can't trust is worse than no briefing — this is the core feature, not a disclaimer.

MODELS

local-first: runs fully offline with ollama (measured usable on a CPU-only box with qwen3:1.7b, hard fallback gemma3:1b). optional frontier layer — claude code subscription token, anthropic api, or any openai-style endpoint — tried first and failing silently down the chain, so a dead key degrades quality, never availability.

DEPENDENCIES

python stdlib. that's it. nothing to pip install, nothing to rot.

SOURCE

github.com/nhantour/nightbrain — MIT.

runs nightly in production at the lab since 2026-07 — the operator reads its output with coffee every morning. dogfooding is the changelog.