LFM 3 is here. • LFM 3 1.7B: reasons before it answers, in 16 languages. • LFM 3 VL 2B: ask about your photos. • The Think switch turns LFM 3 reasoning on or off. • Steadier chats: a stuck answer stops on its own and offers a retry, and earlier reasoning no longer slows the next question. • Fixes for mixed-language conversations and voice input.

Steadier when memory runs low • When an answer gets stuck because the phone is out of memory, the app stops it and offers a fix: try again or pick another model in a new chat • Dictation no longer holds on to memory after you stop speaking, so the next answer runs at full speed • A photo sent without a question is now described • The rating prompt waits for a good answer and never appears on first launch • Stability fixes

Offline loading and improved stability • A new way to pick your first model • Downloaded models load offline • Ready for iOS 27 • Photos are offered only to models that can see them • Type or dictate while the model is still starting • Fewer repeated and looping answers • Answers no longer flicker as they finish

Fixes a crash on launch on iOS 27

Private web search, smoother chats • Private web search with sources • Replies wait while you switch apps • Answers come from the model you pick • Documents and web search in one chat • Dictation that recovers on its own • Model downloads run in the background

Web search, privately • Bring current information into any chat with the new Web toggle • See the pages an answer used, and open any of them • Pages are fetched and read on your device • Watch the search as it happens • More on-device models to choose from

- Chat with your documents, with source citations - Fork conversations, save prompt presets - Faster sending, opening and model switching - Only models your device can run are offered - Response-speed stats are now optional

• Chat with your documents — on-device, with source citations • Fork any conversation to explore a different direction • Save custom system prompts as reusable presets Plus a revamped sidebar with search, smarter model picks, and lots of polish.

• App-wide visual refresh • Model Hub grouped by family with Recommended, Experimental, and Mine tabs • Streaming markdown for assistant replies and thinking blocks • Tap the chat title to rename, export, or delete • Smoother chat input, keyboard, and scroll behavior • Home opens straight into a new chat with your last-used model

- Inline attachments — unified + button in the composer. Vision models get Photo Library / Camera / Document; non-vision models get Document. Attachments render as separate bubbles in chat history. - Per-model system prompts — each model can carry its own default system prompt, applied automatically in new chats. - Bielik models — added Bielik v3.0 and Bielik v3.0 Quantized (Polish-language models). - Expo SDK 55 upgrade. - React Native Executorch 0.9.0 bump.

- VLM support - attach images to messages and use vision-language models (LFM 2.5 VL - 1.6B - Quantized) - Enriched markdown - assistant responses now render with react-native-enriched-markdown for better formatting - Removed Detour SDK - deferred deep link demo integration cleaned up

- react native executorch 0.6.0 - expo sdk 54

- updated detour sdk

- added react-native-detour

- Talk with your documents, on-device RAG - On-Device Speech-To-Text with whisper model - iPads support - Performance improvements

With a v1.0.4 of Private Mind we introduce: - new drawer with better animations and gesture handling - haptics feedback - lots of small improvements and bug fixes

- New splash screen

- About page - Fixed bug when running benchmark overwrote last AI message - Improved process of creating new room by getting rid of screen flicker - Added default system prompt - Correctly attach model to imported chat room