Daemon & Auto-Reflection

Everything you can do by hand, Cerebrumma can do in the background. The daemon runs the file watcher and the dream cycle continuously — so your Brain captures what you work on and distills it into lasting knowledge without you lifting a finger.

Start the daemon #

$ cerebrum daemon start
• Background intelligence started→ watching for file saves + commits→ auto-reflection armed (frequency-capped)

Check on it or stop it any time:

Automatic dream cycle #

The manual dream cycle asks you to paste a reflection prompt into an LLM. With a configured key, Cerebrumma closes that loop itself — reading your episodic memory, running the reflection, and promoting the insights into the right layers.

The daemon runs this same flow on its own once enough fresh episodic memory has accumulated.

Configure a key #

Auto-reflection needs an LLM key. It's stored in ~/.cerebrum/config.json in your home directory — never in your repo— so it's never committed or shared.

config get shows your current settings with keys masked. Without a key, Cerebrumma stays fully offline and the dream cycle stays manual — nothing leaves your machine.

The frequency cap #

Autonomous reflection is deliberately conservative. The daemon only reflects when there's enough newepisodic memory to be worth it, and it enforces a cooldown between runs — so it can never spam your LLM provider or run up a bill in the background. Reflections that only rearrange existing knowledge don't count toward the threshold.

Safety #

Auto-reflection follows the same rules as the manual dream cycle: it promotes insights into semantic, procedural, and personal memory, but never deletes anything automatically. Prune suggestions are always left for you to review.