### What's New - Servers protected by auth proxies (Authelia, Authentik, Keycloak, oauth2-proxy, etc.) now show a sign-in WebView instead of a "proxy authentication" error, letting you authenticate through whatever portal your server uses. ### Improvements - Welcome screen prompt suggestions are now sourced from the server's. - Allow STT language change within the call feature. - Tapping an audio attachment chip after on-device transcription now opens a preview sheet showing the full transcript text, with a copy button, before sending. - Audio files attached in chat are now uploaded to the server by default (server handles transcription automatically). On-Device transcription is available as an alternative in Settings → Speech-to-Text → Audio File Transcription. ### Bug Fixes - File attachments now show a distinct "Processing…" spinner for server-side transcription as well after upload completes, so you can see when the server is indexing or transcribing the file before it's ready to send. - Fixed audio file transcription being cancelled when navigating away from a chat — transcription now continues in the background and completes even if you switch chats. - Fixed channels incorrectly showing as read-only for the channel owner and members with write access when access grants were present. - Fixed background chat content (e.g. prompt cards) being accidentally tappable or scrollable while swiping to open the left or right drawer panel. - Fixed function calling mode not being respected — the app was incorrectly overriding the server's per-model setting; it now lets the server control this entirely, matching the web client behavior.

### What's New - Added Channels — collaborative, topic-based chat rooms where multiple users and AI models interact. - Added Accessibility settings with customizable text scaling — independently adjust message text, conversation titles, and UI elements (buttons, icons, spacing) with live preview and quick presets. - Added slash command prompt library — type `/` in the chat input to browse and search your Open WebUI prompt library. ### Improvements - Inline source citations now appear as small, elevated pill badges showing shortened page titles or domain names — matching the Open WebUI web interface style. - Profile/model avatars will now show properly. ### Bug Fixes - Fixed repeated `heartbeat() missing 1 required positional argument: 'data'` errors in Open WebUI server logs - Fixed web search, image generation, and code interpreter toggles being ignored when turned off mid-chat — toggling a tool off now correctly prevents it from being used. - Fixed conversations older than "This Month" not loading — pagination now properly triggers when scrolling to the bottom, allowing all conversation history to load.

### What's New - Added `@` model mention — type `@` in the chat input to quickly switch which model handles your message. Pick a model from the fluent popup, and a persistent chip appears in the composer showing the active override. The override stays until you dismiss it or pick a different model, letting you freely switch between models mid-conversation without changing the chat's default. - Added Open Terminal integration — enable terminal access for AI models directly from the chat input pill, giving the model the ability to run commands, manage files, and interact with a real Linux environment. - Added Terminal File Browser — swipe from the right edge to open a slide-over file panel. - Added native SVG rendering in chat messages - Added native Mermaid diagram rendering in chat messages (flowcharts, state, sequence, class, and ER diagrams rendered as beautiful images). - Added Memories management (Settings → Personalization → Memories) — view, add, edit, and delete AI memories that persist across conversations. - Added "Archive All Chats" option in the chat list menu for bulk archiving. ### Improvements - iPad layout overhaul - Sidebar drawer now slides smoothly with your finger - Returning to an existing chat now remembers the last model used in that conversation instead of reverting to the default model. - Unified TTS and STT under a single mlx-audio-swift package, replacing two separate dependencies for smaller app size and easier maintenance. - Improved audio transcription for long files with energy-based silence detection for smarter chunking at natural pauses. - User-attached images and files now display inline inside the message bubble instead of floating above it. ### Bug Fixes - Fixed excessive memory usage (1GB+) in long conversations caused by a windowing system that never actually freed off-screen messages due to a self-defeating freeze condition. Replaced with a simple always-render approach which is faster and stutter-free. - Fixed chat search using wrong query parameter, which could cause search to silently fail on some server versions. - Fixed tag removal using incorrect API endpoint format (path-based instead of body-based DELETE). - Fixed tag addition using wrong request body field name. - Fixed tags list fetching from wrong endpoint, now uses the correct structured tags API. - Fixed clone conversation not sending required request body. - Fixed feature toggles (Web Search, Image Generation, Code Interpreter) still appearing in the tools menu even when the admin disabled the capability on the model. Toggles now respect per-model capabilities. - Fixed tool-generated file download links opening in Safari instead of downloading within the app. Files are now downloaded and presented via the share sheet. - Fixed some chats created from the app appearing blank or corrupted on the Open WebUI web interface. - Fixed uploaded photos, PDFs, and other files not displaying on the Open WebUI web interface when sent from the app. - Fixed chat view becoming pannable in all directions after follow-up suggestions appear, instead of strictly vertical scrolling. - Fixed image uploads exceeding the 5 MB API limit by automatically downsampling photos to 2 megapixels before upload. - Fixed external response stream not stopping when clicking the stop button. - Fixed dollar amounts being incorrectly rendered as math equations instead of plain text. - false proxy error should be fixed