Still: a diary that writes itself, as fast as my Raspberry Pi can think

Description and use

The "Still" system auto‑writes a diary on a Raspberry Pi using a local Gemma‑4‑E2B model. Entries are generated only when something noteworthy happens, with silent ticks recorded otherwise. No cloud or API is needed, and hardware limits are respected.

Problem

Automate a personal diary that writes itself only when something noteworthy happens, without external APIs or cloud services, and respects the hardware limits of a Raspberry Pi.

Outcome

The diary now writes only when something noteworthy occurs, producing concise first‑person entries. Silent ticks are recorded, and the system respects the Pi’s hardware limits, avoiding over‑generation.

Workflow

• 1. Run Gemma‑4‑E2B locally via llama‑server on the Pi.
• 2. A worker schedules a tick every real hour, starting after the previous tick finishes.
• 3. The worker queries the Pi’s load average from a separate process to adjust legal moves.
• 4. The worker gathers open items and the current moment, sends them to Gemma for a numbered action.
• 5. Python validates the action against a JSON schema; if invalid, the tick is marked silent.
• 6. Valid actions are executed (e.g., moving parcels, updating files).
• 7. The tick is stored in SQLite; triggers prevent edits or deletes.
• 8. Entries are filtered by a rejection list (sentence count, voice, etc.) before publishing.

Evidence

No API, no cloud, no internet required.

Source

A worker that finishes a tick, then rests one real hour.

Source

Gemma never invents anything. It picks a numbered action from a JSON schema containing only the legal moves, and code validates the answer again.

Source

SQLite as the source of truth. Every tick is stored, including the ones where nothing happened.

Source