Mostly health checks — SPARK is running well. Two commits landed before this session:
119e51e— network-outage resilience: host failure cache + HA offline flag inmind.pyd70403e— README test count correction
a9e997b — two decisions made and committed clean:
- LLM tier reorder (
mind.py): M5 Ollama is now primary for all personas including SPARK. Claude Haiku is the fallback when M5 is unreachable. UsePX_MIND_BACKEND=claudeto force Claude primary. - Expression cooldown (
spark_config.py): raised from 120s → 1800s. One spontaneous utterance per 30 min max.
| Check | Value |
|---|---|
| Health | OK |
| Mood | content (salience 0.85–0.95) |
| Feed posts | 100 (posting actively) |
| Battery | 96%, not charging |
| CPU temp | 53–58°C |
| Disk | 78% — worth watching |
| Claude budget | 3 used / 8 today |
- #36 (MCP server): Phase 1 done. Layers 2–3 not started.
- #25 (SPARK persona): Cognitive loop solid. Obi-facing interaction layer is the remaining gap.
- Commit the pending changes (or make the decisions above and commit clean)
- Disk cleanup — 78% and climbing;
state/thought-images/has 30-day TTL but check if anything else is accumulating - Expression cooldown tuning — live with 30 min for a day and see how it feels
- Issue #36 Layer 2 — stateful conversation buffer. Rolling 10-turn window injected into
build_model_prompt(). Would fix web UI chat statefulness too. ~2–3h. - Issue #25 Obi layer — proximity greet routine, one-question-at-a-time interaction mode, named persona. Needs design before code.
- Issue #31 (face follow) — prerequisite for gaze-toward-Obi on approach
- Issue #36 Layer 3 (autonomous agent) — blocked on Layer 2