What's New - Added dedicated Input Box Text size slider in Accessibility settings - independently scale the font in the chat and channel message composer. - A complete re-write of the streaming pipeline - Massively improved streaming performance. App should feel fully responsive while streaming. Improvements - Switching between chats should feel smoother. - Screen now stays on while TTS is reading a response aloud, just like during voice calls. - Voice calls now respect the admin-configured Voice Mode prompt from OpenWebUI's interface settings - the model automatically receives the right system prompt to keep responses concise and speech-friendly. Bug Fixes - Fixed nested code blocks rendering as plain text. - Fixed citation source icons showing as letter avatars instead of favicons. - Known Issue: Inline visualizer plugin is not correctly working with the new re-write and needs a bit more work. Continue using the native visualizer which works perfectly and much better than pluggin. In future, the pluggin support may be dropped as it requires lot of processing since its written for the webui (includes iframe code in the tool block) and causes lag and has no real benifit over the native visualization.

What's New - Added user valve editor for tools — tap the gear icon next to any tool in the tools picker to configure its settings. - Added in-app update notice — the app checks for new versions on every launch and shows a sheet with release notes; tapping "Later" dismisses it and leaves an update icon next to the New Chat button in sidebar so you can reopen it anytime. - Added star button to every tool in the tools picker — tap the star to instantly pin it as a quick-action pill in the chat input, no settings detour required. Improvements - Tool call results now display with full syntax highlighting and virtual windowing — only the visible portion is rendered, matching how regular code blocks work. - Streaming responses now scroll more smoothly with less jank, reduced CPU usage, and no animation stutter. - Dramatically improved performance for reasoning/thinking models — the app no longer gets slower as the model thinks longer. - Channel deletes, edits, reactions, and pins from other devices now appear live without needing to refresh. - Deleting a channel message now animates out instantly instead of waiting for the server. Bug Fixes - Fixed TTS audio not playing through earbuds or wired headphones in both chat read-aloud and voice calls. - Fixed regenerated responses not showing server errors in the chat bubble. - Fixed the chat not scrolling to the regenerating message — regenerating a response now animates the view to the new message the same way sending does. - Fixed non-toggleable filters incorrectly appearing under Default Filters in the model editor. - Fixed toggle-filter functions in the tools picker always turning on when globally enabled, ignoring the model's configured default state. - Fixed crash in channels when deleting a message. - Fixed channel context menu appearing off-screen when the keyboard was open. - Fixed new chats always starting with the last-used model instead of the server-configured default model. - Fixed Settings Default Model changes not applying to new chats opened in the same session. - Hidden models (disabled by an admin in OpenWebUI) no longer appear in the model picker or the Default Model setting. - Fixed thinking blocks breaking the "Explored N" tool-call grouping on reasoning models — all tool calls now collapse into a single pill as expected.

Drastically improved app’s performance while streaming!

Streaming bug fix

### What's New - Added Native Inline Live Visualization — charts, graphs, svgs and interactive visualizations now render LIVE directly inside chat messages. No need for Inline visualizer pluggin (Works with it as well!). Add "output the code in one file" (or create a prompt in your worspace instead of typing it each time) at the end of your prompt and watch it get built live! ### Improvements - Consecutive tool calls from different tools now group into a single collapsible row, matching the web UI. - Citation badges in AI responses now show domain names by default instead of full page titles, and correctly render grouped citations like [1, 2, 3]. Toggle between domain and title in Settings → Chat Behavior. - Improved math rendering accuracy — formulas inside code blocks now correctly restore their original delimiters instead of showing placeholder text. - Tables with clickable links now handle taps more reliably, and table cells are reused more efficiently for smoother scrolling. ### Bug Fixes - Fixed JavaScript not executing in HTML code block previews — interactive apps like Kanban boards, games, and dashboards now work correctly, including drag-and-drop, button clicks, and localStorage persistence. - Switching accounts now instantly clears the chat list and reloads conversations, folders, and channels for the new account. - Fixed some action buttons not working that required js to complete its task. - Fixed certain tool call Rich UI embeds (music players, video players, dashboards) being non-interactive - Fixed AI message content being clipped at the bottom. - Fixed quick action pills disappearing from the chat input bar after starting a new chat or switching conversations. - Fixed built-in tools (web search, image generation, code interpreter) resetting to their model defaults after sending a message, ignoring any toggles the user had changed.

### Improvements - Significantly improved text streaming — new typewriter-style streaming instead of batches. - small performance boosts when streaming. ### Bug Fixes - Fixed server-side filter content not appearing in the chat until navigating away and back. - Fixed models not able to see image attachments.

### What's New - Added Calendar — view, create, and delete events from your Open WebUI calendars (Personal and Scheduled Tasks). Color-coded month grid with event dots, day event list, create event form with calendar picker, date/time, location, reminder options, and description. Access from the ••• menu in the sidebar. - Added Automations — schedule prompts to run automatically on a recurring schedule (hourly, daily, weekly, monthly, or custom RRULE). Create, edit, enable/disable, run immediately, and view execution history. Access from the ••• menu in the sidebar. - Admin Console → General Settings → Features now includes Calendar and Automations toggles. - Model editor (workspace and admin) now shows Task Management, Automations, and Calendar checkboxes in the Built-in Tools section. - Added task list panel above the chat input — when a model uses task management tools, a collapsible panel shows all tasks with their status. ### Improvements - Screen stays on during voice calls — the display no longer turns off mid-conversation, keeping the call active without needing to tap the screen. - Tool call OUTPUT now shows pretty-printed, expanded JSON instead of a single compressed line. - Multiple consecutive tool calls with the same name are now grouped into a single collapsible row. - Admin users can now open the Admin Console directly from the sidebar (•••) menu. - Various UI element consistency across the app. ### Bug Fixes - Fixed sidebar and chat features (Notes, Channels, Folders, Memories) now correctly respect the user's individual permissions from the server - Fixed message versioning - completely rebuilt end-to-end to match OpenWebUI's conversation tree - Attempting to fix Microsoft (and other OAuth) sign-in staying stuck on the web page after successful login instead of returning to the app.

- | CRITICAL | Open WebUI introduced a new update, 0.9.1, which brings a breaking change related to user messages due to api changes. + Fixed user messages disappearing when re-opening a chat (updated the completions request to include the `user_message` field required by the new server API). ### Improvements - Significantly improved sidebar performance for users with large conversation lists. ### Bug Fixes - Fixed error messages in chat being truncated — full error text now displays without a line limit. - Fixed tapping Photo in the attachment menu returning to the + tools sheet after selecting a photo — the sheet now dismisses immediately when a photo is picked.

### What's New - Added system variables - System variables (`{{USER_LOCATION}}`, `{{USER_NAME}}`, `{{CURRENT_DATETIME}}`, etc.) will now automatically be replaced with their value at runtime matching the webui behavior. ### Bug Fixes - Fixed GPS location not using the device's actual GPS — location is now always fresh and includes a full reverse-geocoded address. Also fixed the chat hanging intermittently when location sharing is enabled.

### What's New - Added GPS location sharing — enable "Share Location" in Privacy & Security settings to send your real location to the AI model when using `{{USER_LOCATION}}` in system prompt for tools like maps, weather, etc. ### Improvements - Redesigned tool call display to provide full details, including rich result items shown inline under each status step. - Tool call status indicator also improved. ### Bug Fixes - Removed further throttling for streaming token by token. - Fixed tool call status history (web search steps, location resolving, etc.) disappearing when switching chats — status updates now persist correctly when reopening a conversation.

### What's New - Added Groups management under user tab in Admin Console — create, edit, and delete user groups, manage group members, configure permissions per group, and set default permissions for all users. ### Improvements - Reorganised Admin Console into cleaner top-level tabs (Users, Analytics, Functions, Settings), with Settings containing a searchable section picker for General, Connections, Integrations, Documents, Web Search, Code Execution, Interface, Audio, and Images. ### Bug Fixes - Fixed code blocks flickering/flashing colors during streaming - Fixed HTML code blocks not rendering as live previews after the AI finishes responding. - Fixed thinking/reasoning block staying expanded after thinking completes — it now collapses automatically once the model finishes reasoning. - Fixed profile avatar related issues.

### Bug Fixes - Fixed TTS mispronouncing numbers

### What's New - Admin's Paradise (Will bring the ones left in further releases): - Added Analytics dashboard, General, Connections, Integrations, Code Execution, Documents, Web Search, Interface, Audio, and Images settings to admin console with full control. - Added multi-account support per server — login to multiple accounts on the same server and switch between them instantly with a long-press on your avatar in sidebar or from the setting -> manage servers. - Added robust automatic server reconnection and a user facing error message when unable to reach the server. - Added Python code execution — tap the Run button on any Python code block to execute it on-device using Pyodide (WebAssembly). Supports numpy, pandas, matplotlib, sympy, and more. ### Improvements - New tools now start with a helpful Python template - Images in AI responses now render inline — markdown image links display as actual images, tap to open the linked page. - Redesigned the Account page in Settings - Added ability to modify account details. - Added group-based access control — you can now grant access to entire groups (not just individual users) across Prompts, Knowledge, Models, Tools, Skills, and Channels. - Tapping a processed file attachment now shows the extracted text content and a native PDF preview in a tabbed sheet, matching the web UI experience. - Streaming responses now grow smoothly per-character without visible height-update chunking. - Extended tool API request timeout to 5 minutes to prevent timeouts when loading or saving complex tools. - Added a tip below the TTS voice preview button explaining that it lets you hear how the selected voice sounds. ### Bug Fixes - Fixed follow-up suggestions being cut off after two lines — they now expand to show the full text. - Fixed Voice Call not transcribing speech when connected to CarPlay — the microphone now works correctly through the car's hands-free system. - Fixed deleting a chat, channel, or folder chat leaving the screen stuck on stale content — the app now automatically navigates to the new chat screen after any deletion. - Fixed cloning tools failing due to wrong character in name. - Fixed saving workspace items (tools, skills, prompts, models) showing a false "session expired" error.

### Improvements - Reverted streaming responses back to appearing character-by-character instead of arriving in chunks - Feels smoother! ### Bug Fixes - Fix scrolling to bottom when entering chats. - Fixed background notifications not reliably delivering

### What's New - Added "Show Response Preview" option in Settings → Notifications — when enabled, the first lines of the AI response appear in the completion notification. Off by default. ### Improvements - Significant Markdown library performance updates. Increased smoothness while streaming long responses! - User avatars are not fetched at app restart instead of using stale data. ### Bug Fixes - Fixed Qwen3 TTS switching to an English accent mid-response when speaking non-English text — the voice now stays in the correct language. - Fixed Qwen3 TTS producing corrupted/distorted audio during voice calls after a few sentences due to unbounded GPU memory growth — memory is now properly cleared after each sentence. - Fixed voice call TTS continuing to speak the full response after disconnecting — ending a call now stops playback immediately. - Fixed voice call TTS randomly pausing and skipping sentences — the audio pipeline now stays open for the full response instead of tearing down between sentences. - Fixed the same response being spoken twice during voice calls — the streaming and speak pipelines no longer overlap. - Fixed switching from a voice call to chat read-aloud causing audio glitches — each session now cleanly resets state before starting.

### What's New - Added per-chat advanced parameters panel — tap the sliders icon in any chat to override temperature, top-p, system prompt, and 20+ other parameters for that specific conversation. - Added Reference Chats — include a previous conversation as context in any message via the + attachment menu. - Replaced Marvis TTS with: - Qwen3 TTS (Recomended) — supporting English, Chinese, Korean, Japanese, German, Spanish, French, Italian, Portuguese, Russian, and Arabic, with multiple speaker voices and language selection in Settings. - Added Kokoro TTS as a secondary option — 54 voices across 9 languages (American/British English, Spanish, French, Hindi, Italian, Japanese, Portuguese, Chinese) with adjustable speed. ### Improvements - Corrected slider ranges and states in the workspace model editor. ### Bug Fixes - Fixed tapping the edit button in the model picker not opening the model editor for certain models. - Fixed saving a new workspace model showing a false "session expired" error. - Model picker no longer floods the server with hundreds of simultaneous image requests during fast scrolling - Fixed prompt library, knowledge, model, and skill pickers going behind the navigation bar when using a large third-party keyboard. - Fixed pinning a model in model selector would not update the star immediately. - System prompt is now also sent in request params for better compatibility with server-side prompt handling.

### What's New - Redesigned appstore screenshots and the icon! ### Improvements - Added haptic feedback when tapping the model selector. ### Bug Fixes - Base model picker when creating a workspace model now only shows provider models, not other workspace models. - Model ID, Prompt command, Skill ID, and Tool ID fields now correctly auto-fill from the name when creating new items in the workspace. - Added consistent pill background to the navigation bar model selector chip.

### What's New - Enabled Mac (Designed for iPad) destination for the app - Allows you to download the iPad version of the app on your mac. ### Improvements - Reduced memory usage for large conversations! - Big Channels UI improvements. - Scrolling up during a streaming response now breaks out of auto-scroll immediately. ### Bug Fixes - Fixed "Send on Enter" not working in channels and thread replies — pressing Enter now sends the message as expected when the setting is enabled. - Fixed thread replies appearing twice in the thread view when sent

### What's New - Added inline #URL detection — type `#` followed by a URL in the chat input to see a suggestion pill; tap it to scrape and attach the webpage as a file. - Added tap-to-preview for all attachment types — tap any image, audio, or file pill in the input bar to see a fullscreen preview. ### Improvements - Sending a message now smoothly scrolls your question to the top of the screen so the AI response streams in below it. - Reverted back to orginal scrolling behavior until further polishing. The memory will still be significantly lower if code blocks are in the chat. - Adding a website URL via the + button now scrapes the page and attaches it as a file instead of pasting the URL into the text box. - Using the prompt library with "/" now appends the selected prompt to your existing text instead of replacing it. ### Bug Fixes - Fixed knowledge, prompt, skill, and model picker overlays covering the text input field — the input box now stays visible when any picker is open.

### What's New - Added voice dictation — tap the mic button in the chat input bar to dictate, then tap Stop to append the transcribed text to your message using on-device or server model. - Full prompt versioning support - Version history for every message is properly preserved where previously on worked for assistant messages. ### Improvements - Drastically reduced cpu/memory usage across the app - About 70%+ (scaled with #/length of messages) drop in memory and 20-40% in cpu utilization for same tasks. - Tapping "Edit" on a channel message now opens the keyboard automatically so you can start typing right away. - The thread replies sheet now opens at a comfortable near-full-height and can be dragged to dismiss. ### Bug Fixes - Fixed channel unread badges never clearing — opening a channel now marks it as read and clears the badge immediately. - Fixed channels and thread replies ignoring the "Send on Enter" toggle — pressing Return now correctly inserts a new line when the toggle is off. - Fixed channel reactions added from the web showing as raw shortcode text (e.g. "sunglasses") instead of the actual emoji. - Fixed thinking blocks swallowing the model's actual reply when the model omits the opening think tag — the response now renders correctly below the collapsed reasoning block.

### What's New - Experimental significant memory reduction approach + a more responsive UI and faster streaming. ### Bug Fixes - Fixed Shift+Enter not inserting a new line on the first use after app restart on iPad with a hardware keyboard.

### What's New - Added Storage browser in Settings — view all app storage usage and use quick-action buttons to clear caches or remove ML model files in one tap. - Added multi-language support for 56 languages (Hindi, Chinese, French, German, Japanese, Korean, Spanish, Polish, and many more). - Added in-app Language picker in Settings → Display — browse all supported languages. ### Improvements - Welcome screen prompt cards now prioritize per-model suggestions over global admin prompts — model-specific prompts show first, with admin-configured prompts as the fallback. - The server connection screen, onboarding, and About screen now display the actual app icon instead of a generic placeholder icon. - Long-press any pinned model in the sidebar to unpin it directly, without having to open the model picker. ### Bug Fixes - Fixed Marvis Neural TTS producing garbled/garbage audio on responses that contain bullet lists, or paragraphs ending with a colon — the text preprocessor no longer generates invalid "colon-period" sequences that the model can't handle. - Fixed server-side TTS accumulating gigabytes of temporary audio files over time — each spoken sentence now deletes its temp file immediately after playback, and any unplayed files are cleaned up when TTS is stopped. - Fixed app crash when backgrounding during on-device speech-to-text transcription. - Fixed thinking/reasoning blocks not responding to taps while a response is streaming — you can now expand or collapse the thinking block at any time during streaming. - Fixed memories getting disabled by itself — pinning a model, changing the default model, or toggling memory from any screen no longer wipes other user settings. - Fixed default model not sticking — pinning a model was incorrectly overwriting the default model setting with the pinned models list.

### What's New - Added German and French conversational voice options for Marvis Neural TTS. - Added minimize/PiP for voice calls — tap the chevron button in the voice call screen to shrink it to a floating pill. Tap the pill to restore the full call, or tap the red button to end it. The call stays active while minimized. ### Improvements - Drastically improved TTS text-to-speech naturalness: Sentences now create proper pauses, even better than openwebui splitting. ### Bug Fixes - Fixed voice calls not starting to speak until the full AI response finished generating — responses now begin playing as soon as the first complete sentence arrives. - Fixed audible gaps between spoken sentences in server-side TTS — replaced the old polling-based audio player with gapless queue playback so chunks play back-to-back without any pauses.

### What's New - Added Action Buttons support ### Improvements - Replaced Parakeet (English-only) with Qwen3 ASR for on-device audio transcription — now supports automatic language detection and multilingual transcription (Spanish, French, German, Italian, Portuguese, Russian, Chinese, Japanese, Korean, and more). - Model editor now supports enabling/disabling action buttons - Toggle-filter functions now appear as toggleable tools in the Tools menu alongside regular tools. - Filter functions are now properly resolved using the global vs per-model logic — global filters always apply, per-model filters respect configuration. - Starter prompt cards on the welcome screen now fall back to per-model suggestion prompts when the admin hasn't set global prompts, and update automatically when switching models. - The TTS/STT settings screen now correctly shows "Not Loaded" (when the model is downloaded but not in memory) vs "Not Downloaded" (when no model files exist on disk), and the download/load button label and icon also adapt accordingly. ### Bug Fixes - Fixed on-device TTS and STT models taking up twice the expected storage — the HuggingFace download library was leaving a duplicate blob cache alongside the working model files. Existing users will automatically reclaim the wasted space on their next app launch. - Fixed Shift+Enter intermittently sending the message instead of inserting a new line on iPad with a hardware keyboard. - Fixed accessibility sizing not applying to assistant messages, drawer lists, and input boxes. - Fixed orphaned think closing tags leaking into chat messages as visible code blocks when models like Qwen skip the opening tag or when streaming splits tags across chunks. - Fixed selecting a model and immediately sending a message no longer uses stale config.

### What's New - Added pinned models - star any model in the model picker to pin it for quick access. Pinned models appear in a dedicated section at the top of the picker and as shortcuts in the sidebar, synced with your Open WebUI server. - Model picker now shows the currently selected model at the top of the sheet for easy reference. ### Improvements - home screen widgets with full theme support — widgets now properly adapt to Default, Dark, Clear, and Tinted modes instead of being stuck on a dark background. ### Bug Fixes - Fixed thinking/reasoning blocks from models (Qwen, DeepSeek, etc.) showing as raw tags in the chat instead of rendering as a collapsible "Thinking" section. Now handles all reasoning tag formats during streaming and fixes stray tags leaking mid-stream. - Fixed on-device audio transcription cutting off the last portion (and many words throughout) of uploaded audio.