v0.0.6
Highlights
Cost controls
- Per-task model tiering — one base model for chat/autopilot, optional cheaper mini/nano tiers for the small background jobs.
openagi modelsshows the plan + savings. Configure withOPENAI_MODEL_MINI/NANO(orANTHROPIC_*) + per-task pins. - Autopilot routes through tiering and has a tunable cadence (
OPENAGI_AUTOPILOT_INTERVAL_MIN) — an idle agent can go from ~$40/day to a couple dollars. - The budget ledger now records the actual model used per call.
iMessage (macOS node)
- Note-to-self — text yourself to command the agent.
- Whole-word keyword trigger — say "Peri …" to invoke; everything else is silently captured.
- Group chats — in an allowlisted group, any member can invoke the agent and the reply posts back to the group (
--allow-chat). - Proactive extraction — a cheap nano pass turns incoming texts into follow-up tasks, plus events and links in memory.
- Stability: reads
chat.dbstrictly read-only (no longer wedges Messages) and tracks new messages by date, surviving Messages' database rebuilds.
Reliability
- Self-update auto-recovers from upstream history rewrites — a clean checkout whose history was rewritten (e.g. a force-push) hard-resets onto upstream instead of silently failing; real local commits and uncommitted edits are never discarded.
main. Auto-update self-heals on a clean checkout (see above). If yours has local edits, run git reset --hard origin/main once.