Local voice chat is here: - Record from the composer and have your message transcribed on device. - Download Moonshine Base for private local speech-to-text. - Audio transcripts are saved with messages and can be expanded in chat. - Replies can be spoken aloud after a voice prompt. - Improved Cactus model selection, context limits, and prompt-cache handling for more reliable first responses.

-Refreshed settings screens with a cleaner Liquid Glass design. -Smarter token defaults and response limits for local models. -Improved model selection behavior with better per-model recommendations. -Added Cactus dSYM support for better crash diagnostics. -CI and build reliability improvements.

This update improves ride sharing for larger exports. You can now more easily save full ride videos to Files, use AirDrop or cloud storage links, and create shareable versions for friends, family, or social posts. We’ve also polished the export flow to make sharing smoother and more reliable.

This release refines the chat composer with a cleaner layout, improved attachment and action controls, and better spacing around the input area. Sent messages now appear immediately before model warm-up completes, making first sends feel more responsive and avoiding awkward empty-state transitions.

FoundationChat 1.0.2 adds smarter local tool use for supported on-device models. - Models can now use local tools for date and time, calculations, unit conversions, conversation lookup, attachment context, and model/session status. - Improved Foundation Models integration with native tool support. - Added tool-aware support for compatible Cactus models. - Refined Cactus warm-up behavior for more reliable first responses. - Continued privacy-first design: conversations, attachments, and model interactions stay on your device.

• New & updated downloadable models — Expanded lineup of on-device weights you can pull in and run locally, with clearer choices in model management. • Stronger local inference — Improvements to the on-device inference engine for loading, running, and switching models more reliably. • Downloads & setup — Smoother model downloads and fewer rough edges when adding or replacing models. • Reliability — Better behavior with large models and long, streaming replies, including more dependable saving while a response is still generating. • Polish — Bug fixes and small UI improvements.