Styles for pictures • Pick a style before you make a picture: Photoreal, Cinematic, Anime, Sketch, Watercolor, Oil painting and more. Arrows show you there are more styles to see. • Tap Restyle on any picture you made to try it in another style. • Pro: clips get styles too. Clearer video clips on iPhone • The default video model now draws a clear scene instead of colour blobs. Clearer messages • Search and video models say why they did not load. Quick settings now fit the screen on iPhone.

Make videos on your iPhone Describe a short clip and Off Grid AI makes it on your phone. Nothing is uploaded. Watch it in the new Videos tab, save it, share it, or play it picture-in-picture. New models • LFM3 1.7B for fast, sharp answers. • LFM3 VL 2B reads photos, documents and charts. Better pictures • Pictures render properly on iPhone. No more blank white images. • Use full Stable Diffusion checkpoints and Qwen image packs, and find more on Hugging Face from inside the app. • Watch your picture form while it's being made. • Each image model keeps its own steps, guidance and size. Voice and dictation • Faster, more reliable dictation and spoken replies. • Your words are kept even if something goes wrong mid-recording. Easier to get started • Auto Setup picks models that fit your phone, including video and speech when there's room. Fixes and polish • Stop keeps what was already written, and stops a video sooner. • Clear messages when a model can't load or a picture can't be made, with Load Anyway when memory is tight. • Deleting a chat also removes its pictures and videos. • Downloads keep their progress and pick up where they left off.

Bigger models on your iPhone. A 12 GB iPhone now runs 9B models. No "Load Anyway" needed. The model list and picks now match what your iPhone can actually run. One tap, the whole model. Split models, vision files and speed add-ons download together on one progress bar. No parts to track down. Faster replies. A model with a speed add-on gets it on its own. New badges on each model card show which models answer faster, and why. Fastest settings, up front. "Use fastest settings" now sits at the top of model settings. Search that stays with you. The search bar stays put in Chats and Models while you scroll. Projects now has search too. Drag the list to put the keyboard away. New models, no update. New models now reach your phone without an app update. VoiceOver now reads what each speed badge means. Everything still runs on your phone. Nothing leaves it.

NEW IN THIS VERSION A model that fits your iPhone, in one tap The Models, Image and Speech tabs now open with three picks made for your iPhone: Fast, Balanced and Best. Tap one. It downloads, and you can start chatting as soon as it is ready. Keep every answer Retry, resend or edit a message, and the earlier answers stay. Go back to any of them. Line up your next questions Type while a reply is still running. Your messages wait in a queue. Send one now, or remove it. Search shows only what runs Models that are too big for your iPhone, or that this version cannot load, are hidden. Each model shows once, with its sizes in plain words: Smaller, Recommended and Best. New models without an app update Tap "Check for new models" to add models as soon as we publish them. The fastest settings for your iPhone "Use fastest settings" tests each engine your iPhone can run and keeps the fastest. The test takes 30 seconds or less. FIXED • Switch apps during a reply, and the model keeps working when you come back. • If a new model does not fit in memory, your current model stays loaded. • On a nearly full iPhone, image generation stops with a storage message instead of closing the app. • After a 30-second dictation, the next tap on the mic records again.

Chat and model loading - Text models now load when first used, rather than when selected. - Your chat stays visible while a model loads, without a full-screen loading page. Fixes - License activation stays reachable when the keyboard is open. - The speech model list inside chat now scrolls. Image generation - New image settings default to 512 x 512. - Removed the suggestion to switch to 256 x 256. Your saved settings stay unchanged.

• Pair with Desktop by QR code, then use Computer Use, Web Use, and Assistant from Mobile. • Turn Assistant on or off for each message and follow live Desktop tasks in chat. • Search chats and delete multiple conversations at once. • Pause and resume supported model downloads. • Improved Personal Mesh sync, remote models, image and voice workflows, and chat reliability. • Support Bonsai 2 27B via OGAD

Sync is live. Pair your iPhone with your other Off Grid devices and continue your chats and projects there. Move generated images, attachments, selected files, and compatible installed models between paired devices. Choose whether copied text and new screenshots also move between your devices. Voice conversations now stop when you stop speaking and are easier to interrupt. Model transfers are faster and more reliable, including after an app update or reinstall. Improved model loading, error messages, accessibility, and everyday controls.

This is a big stability and control release. New: - Model Loading modes (Lean, Balanced, Aggressive) to control how much memory models get - Model manager shows what's loaded and its RAM, with per-model Eject and Eject All (voice and transcription included) - Transcription (STT) and Text-to-Speech (TTS) now in Model Settings - Turn Thinking on or off for reasoning models right from chat - Cloud marker distinguishes remote models from on-device ones - Repair action for vision models missing their projector file - "Stay in the loop" card in Settings, plus Follow on X and Join Slack links Fixes & improvements: - Downloads interrupted by an app-kill come back as retryable cards instead of vanishing; queued downloads survive a restart and resume - Failed downloads show in the badge count and Download Manager - Voice notes and dictation transcribe reliably; push-to-talk has a clear slide-to-cancel pill and the mic stays usable while a model loads - Stopping a reply keeps what was already written, reasoning included, and no longer wedges the chat or mislabels a stopped turn; a genuine token-limit cutoff now shows a clear indicator - Thinking toggle applies to the current turn; remote reasoning (LM Studio, Ollama) now shows in the Thinking block - Broken or partly-downloaded models are caught before generation with a plain error instead of a silent failure - 9B models load and run fast; switching models goes through the memory-aware loader so it doesn't run out of RAM - Load Anyway is offered for any model over budget and measures real free memory after clearing space - Image generation guides you on the GPU speed/quality trade-off and no longer produces garbage-resolution output - Image settings use your values, and Reset to Defaults resets them too - Context-length slider goes up to each model's real trained maximum - Knowledge-base search only runs when you turn it on - The image viewer closes before the Save prompt, and audio releases before the photo picker, fixing input and voice-mode hangs - Faster first launch, and Aggressive mode no longer overcommits memory

Aggressive Loading mode for max single-model performance. "Load Anyway" when a model won't fit. Fixes: voice transcripts never send empty, voice image requests route correctly, no more reasoning text leaking into replies, cleaner model cards, and fewer out-of-memory crashes.

Voice Power: Auto-send voice notes and see transcriptions in Chat. Speed Boost: Major performance and stability improvements for a smoother experience. UI Polish: New features and cleaner design elements on the Home screen.

Now with Voice Mode Talk to your AI hands-free, fully on-device. No accounts, no tracking. Voice Mode: speak naturally, hear it answer back. Adjustable speed + tap-to-seek replay. Built-in transcription with multiple Whisper models. Pro: connect email, calendar, and external tools over MCP. Smarter model routing and faster loading. Offline license verification — no account needed. Update and say hello, out loud.

• Introducing a new PRO Experience • Support for Gemma 4 models • Import vision models with mmproj support • Better download recovery and retry handling • Improved document picker and UI modals • Fixed SDXL validation issues • Fixed knowledge base upload handling with iCloud • Added remote model API key support • Performance improvements and bug fixes

Bug fixes, and performance enhancements

- Connect to Ollama, or LM Studio on your network — with automatic LAN discovery - Knowledge Base for projects: attach documents and use them as context in chats - Bug fixes and performance improvements

Qwen 3.5 support You can now run Qwen 3.5 models locally on your device. Smarter conversations You can now send URLs to the AI and it will read and summarize web pages for you. All built-in tools are now enabled by default so you get the full experience right away. Faster responses We've optimized inference speed with dynamic context scaling, so you'll notice quicker replies especially during longer conversations. Image generation fixed on iOS SDXL image generation models now work correctly. Previously the app would crash when using models like Apple's CoreML Stable Diffusion XL. That's sorted out now. Stability improvements Fixed a crash that could occur on older iPhones with limited RAM (like iPhone XS). Also resolved a data race issue in the download manager on iOS.

Your AI just got smarter - and it's still 100% offline. Off Grid now supports tool use: web search, calculator, date/time, and device info — all running on-device with automatic reasoning loops. Ask a question, your AI finds the answer, no cloud required. Also new: configurable KV cache for faster inference, expanded vision models, and dozens of polish and stability improvements. Your AI. Your device. Your data.