Hannah v1.0.10
The brain is validated when you choose it. Key and model are checked against the provider (/models); a retired model (Groq dropped llama-3.1-8b-instant, the old default) is refused with the models your key can use, one click to pick. A model failing mid-conversation reopens the brain screen with the real message instead of a silent turn. Groq default is now llama-3.3-70b-versatile.
Also since v1.0.9
- Listening starts as soon as a brain exists, even when it was configured from the settings panel.
- If the microphone cannot be opened (OS privacy setting), the HUD says so.
- The gesture model warms up in the background;
hannah doctorsays "warming up" instead of "missing" on the first start. - Avatar upload errors show the real cause.
Assets
Unsigned builds from native runners: -linux-x86_64.AppImage + .deb, -mac-arm64.dmg, -mac-x64.dmg, -win-x64.exe, SHA256SUMS.