Skip to content

v0.2.12 — openmuse ships in the repo

Latest

Choose a tag to compare

@github-actions github-actions released this 19 Oct 19:25
· 5 commits to main since this release

v0.2.12 — 2026-10-19 — openmuse ships in the repo

The optional agent layer now lives in this repository (openmuse/) instead of a separate project: one clone gets the firmware patch and the agent, and apply.sh --preset openmuse wires the device to it.

  • In-repo agent — Hono/Node service with an OpenAI-compatible surface (SSE passthrough); Dockerfile/compose, .env.example, and openmuse/run.sh (Docker when available, Node otherwise; --funnel exposes HTTPS via Tailscale).
  • Context — auto-detects the host's city/timezone from its public IP and injects them into every system prompt; weather via open-meteo; "remember/记住" facts persist to disk (openmuse/data/memories.json) or KV on Workers.
  • STT proxy — /v1/audio/transcriptions forwards to the provider, so the device keeps one TLS session for STT + chat.
  • Preset — apply.sh --preset openmuse points chat and STT at the agent and keeps QuickSilver for TTS; it asks only for the agent URL and token.
  • CI — new openmuse job (npm ci + syntax check); openmuse/run.sh is shellchecked.
  • Verified on device — weather with no city (auto San Jose), time, memory across a service restart, Chinese STT through the proxy, and a full voice turn (k self-test) in 4.7 s; typed warm turns ~3.9 s.