Release notes · November 2026
Beta 1.2.2.
More useful Omni conversations and clearer spoken actions.
What changed
- Omni conversations: task-aware guidance works through the shared provider adapter. Response style, personality, and spoken progress can be adjusted. Short same-model follow-ups retain recent context while connected-data restrictions remain in force.
- Spoken actions: short acknowledgements come from actual tool starts. Omni does not speak model-authored plan reasoning. Hidden reasoning and common Markdown are removed before speech.
- Local voice output: the signed speech helper uses macOS AVSpeechSynthesizer and discovers installed voice quality. Automatic output remains on this Mac; available voices can be previewed and speech can be interrupted. Better Enhanced or Premium voices must first be installed through macOS.
- Connected tools: Omni recognizes Calendar and GitHub requests in larger tool catalogs, and a new system-information tool reports the actual macOS version and architecture. A redundant initial planning gate no longer blocks some local-model actions.
Voice and testing limits
The release host had no microphone input device during this work. The user reports that speech recognition works well, so development testing focused on output. The local speech helper completed synthesis and interruption checks, but human listening is still needed to judge naturalness. Only default-quality English voices were available to the signed helper on this host; the resulting voice may still sound synthetic. Install an Enhanced or Premium voice in macOS System Settings → Accessibility → Read & Speak → System voice if offered by your Mac.
ElevenLabs remains an optional cloud voice provider. It sends prepared speech text to ElevenLabs only after you explicitly connect and select it, and usage may cost credits. Its live output was not retested without a configured account. Local Ollama requests exercised real read-only Gmail, Calendar, and GitHub tools; no email was sent or external record changed. AI models can still misstate tool results or ignore exact formatting instructions.
Compatibility and updating
Current supports macOS 15 or newer; Sonoma Legacy supports macOS 14. Both editions have Apple silicon ARM64 and Intel x64 installers. The beta updater can discover and integrity-check a compatible DMG; installation remains manual.
Previous installers, URLs, and checksums remain in the archive. Cross-built Sonoma and Intel artifacts are not described as runtime-tested on macOS 14 or physical Intel hardware.