lucerna 0.1.0

Self-regulating autonomous AI work orchestrator — dispatches LLM agent waves, manages budget headroom, runs cross-model quality review

Lucerna

Greek: ἡγεμών (hēgemṓn) — the organizing intelligence

Self-regulating autonomous AI work orchestrator. Dispatches LLM agent waves, manages budget headroom, and runs cross-model quality review.

What it does

Lucerna runs 24/7 as a daemon, checking for budget headroom every 30 minutes. When surplus exists, it dispatches autonomous AI work waves. When resources thin, it stops — ensuring interactive sessions always have priority.

Safety nets

1. Live budget check before each dispatch cycle
2. Morning deadline — stops dispatching before your workday starts
3. Max waves cap — bounded damage per cycle
4. Circuit breaker — stops after consecutive failures
5. Cross-model quality review — independent models review each wave's output

Cross-model review

After each wave, Lucerna dispatches the manifest to multiple LLM CLI tools (Gemini CLI, Codex CLI) for independent quality review. Only cross-model disagreements surface to the human operator — you're the tiebreaker, not the reviewer.

Status

Alpha — currently tightly coupled to the author's personal AI stack. Generalisation in progress.

License

MIT