What's Changed
- Improved streaming replies in chat, with cleaner completion on terminal chunks, smoother long-conversation behavior, and better handling when you manually scroll during generation.
- Switched reasoning and tool-call rendering to native collapsible markdown blocks, making assistant responses cleaner and more consistent.
- Added richer embed and preview support in messages, including HTML/SVG preview actions and more reliable sizing for WebView-based content such as charts, Mermaid, and other embeds.
- Improved proxy and SSO authentication reliability by queueing JWT capture, retrying delayed token capture, and better handling OAuth flows that need in-WebView sign-in to fully complete. Fixes: #407
- Fixed server persistence for streaming assistant messages so in-progress replies are not marked complete too early, allowing the backend to continue writing to the same branch correctly.
- Improved text-to-speech with backend-driven server voices, default voice loading, and text split modes like punctuation, paragraphs, and no splitting to better match backend behavior. Fixes: #409
- Refined haptic feedback across the app, preferring native Android system haptics while keeping richer notification-style feedback on iOS.
- Clarified knowledge base attachment picking by treating cached entries as files and surfacing clearer file names in the picker. Fixes: #337 and #308
- Added broader regression coverage and stability work across auth, streaming, markdown rendering, haptics, proxy auth, and WebView height measurement.
- Fixed quick actions for cold starts on iOS and Android.
Full Changelog: v2.6.2...v2.6.3