You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This commit was created on GitHub.com and signed with GitHub’s verified signature.
Fixed
Desktop: Local TTS and voice input couldn't reach servers that don't send CORS headers, like macos-speech-server, and showed "Could not reach a local TTS server" even though the server worked from the terminal. The desktop app now sends these requests from its native core, so a local server needs no CORS setup (#317, closes #316).
Local TTS servers that only speak WAV refused every request. Utsuwa now asks local servers for WAV instead of MP3. Kokoro-FastAPI and openedai-speech support both (#317).
A local TTS server with no voice list showed as unreachable in the provider menu. A missing voice list no longer counts as down (#317).
Changed
MCP tools that return images now give the model a short [image: image/png] note instead of the raw image data, which could fill its context. MCP responses over 12 MB are refused. This is the groundwork for picture search, contributed by @dezihh (#307).