Self-host Open-WebUI
v0.13.3Self-hosted AI platform for local and cloud models
- Stars
- 158 683
ghcr.io/open-webui/open-webuiWhat is Open-WebUI?
Open WebUI is a self-hosted AI platform that serves as a unified interface for interacting with large language models. It connects to local models via Ollama and cloud models via OpenAI-compatible APIs. The platform provides a rich chat interface with Retrieval-Augmented Generation backed by nine vector databases, web search across dozens of providers, image generation with multiple engines, voice and video calls with various speech engines, a plugin system for extensibility, persistent memory across conversations, notes, channels for team collaboration, calendars with AI scheduling, and automations. It supports granular RBAC, LDAP, OIDC, and SCIM provisioning for enterprise deployments. Installation options include Docker, pip, Kubernetes, and native desktop apps. The default port is 8080. It uses SQLite by default with optional PostgreSQL, and supports horizontal scaling with Redis. The project uses a custom license requiring branding preservation and offers an enterprise edition.
Should you self-host Open-WebUI?
Open-WebUI scores 90/100 on selfhost.directory (how scoring works) and ranks #3 of 123 in AI. It has 158 683 GitHub stars, +4 354 in the last 30 days.
It shipped 28 stable releases in the past year. The latest, v0.13.3, came out on 8 November 2026. New versions land roughly every 4 days.
Setup is moderate: expect some configuration beyond a single command. An official Docker image is available. It is BSD-3-Clause-licensed and written mostly in Svelte.
Its closest alternative on selfhost.directory is openclaw (93/100, 394 480 stars). It scores higher overall, so compare the two before you pick. See Open-WebUI vs openclaw side by side.
Open-WebUI README
Open WebUI is a home for AI: an extensible, easy-to-use, self-hosted platform built to run entirely offline. Connect local models through Ollama or use OpenAI-compatible APIs for local and cloud models.
English | 简体中文 | 繁體中文 | 한국어 | Deutsch | Español | Français | Italiano | Dansk | 日本語 | Polski | Русский | Bosanski | العربية | Norsk | Português (Brasil) | ไทย | Türkçe | Українська | বাংলা | Ελληνικά | Tiếng Việt
# With uv installed (macOS/Linux) DATA_DIR=~/.open-webui uvx --python 3.12 open-webui@latest serve # Or install and run with pip (Python 3.12) pip install open-webui open-webui serve
Then open http://localhost:8080.
docker run -d -p 3000:8080 --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:main
Then open http://localhost:3000.
| Image tag | Includes |
|---|---|
:main |
Standard image with local AI dependencies |
:slim |
Lightweight image using external services for embeddings, document extraction, and voice |
:cuda |
Open WebUI with CUDA acceleration |
:ollama |
Open WebUI bundled with Ollama |
:dev |
Latest development build |
Add these options to docker run before the image name as needed:
| Option | Purpose |
|---|---|
-e OLLAMA_BASE_URL=http://your-ollama-server:11434 |
Connect to Ollama on another server |
-e OPENAI_API_KEY=your_secret_key |
Set your OpenAI API key |
-e HF_HUB_OFFLINE=1 |
Disable Hugging Face model downloads for offline use |
--gpus all |
Enable Nvidia GPU access for :cuda or :ollama; requires the Nvidia Container Toolkit |
See the Docker installation guide for setup instructions and configuration options.
Download the native desktop app for macOS, Windows, and Linux.
- Open Terminal / Terminals (Enterprise): The action layer that turns Open WebUI into a full agent harness, with enterprise orchestration for per-user workspaces.
- Knowledge Base Sync (oikb): Mirror folders, repos, buckets, and wikis into Open WebUI Knowledge Bases with incremental sync.
- Open WebUI Computer: Your computer in any browser: files, editor, terminal, git, and optional AI agents, running on your real machine.
See the ecosystem documentation for details.
See the troubleshooting guide for connection issues and the updating guide to update your installation.
See the docs for configuration and usage, and the roadmap for planned work.
For Docker Compose, Kubernetes, and other setups, see the installation docs.
Open an issue or start a discussion.
Find us on Reddit, X, and Discord.
Report vulnerabilities privately through GitHub security advisories. Security reports are accepted only through GitHub.
How to configure Open-WebUI
HSTSvalues: max-age=<seconds>, optionally followed by ;includeSubDomains and ;preload (no spaces)PERMISSIONS_POLICYvalues: Comma-separated feature=() or feature=(self) entries without spaces, for accelerometer, autoplay, camera, clipboard-read, clipboard-write, fullscreen, geolocation, gyroscope, magnetometer, microphone, midi, payment, picture-in-picture, sync-xhr, usb and xr-spatial-trackingREFERRER_POLICYvalues: no-referrer, no-referrer-when-downgrade, origin, origin-when-cross-origin, same-origin, strict-origin, strict-origin-when-cross-origin, unsafe-urlXCONTENT_TYPEvalues: nosniff (any value sends nosniff)XDOWNLOAD_OPTIONSvalues: noopen (any value sends noopen)
Open-WebUI releases & version history
Added
- 🦾 Agents in automations. The automation editor now lets you choose a terminal and working folder, then select a coding agent available there. Scheduled runs use that saved terminal and folder, and check that the terminal and agent are still available before starting. Agent automations send their results to a chat; channel destinations remain available for ordinary models. Commit
- 📂 Change terminal and folder during a chat. You can now choose a different terminal or working folder after sending messages, including in temporary chats. Both controls remain disabled while a reply is being generated. Commit
- 🇧🇷 Brazilian Portuguese translations. Added missing translations for terminal and folder controls, Git actions, spreadsheet editing, voice notes, tab shortcuts, notification sounds and tool settings. Updated the explanations for document retrieval, tool permissions and voice settings. #33945
Fixed
- 📂 Files tabs follow terminal changes. Switching a chat to another terminal now updates its open Files tabs to show that terminal's files, instead of leaving the previous terminal's folders on screen. With a terminal set to "Per chat", the Files tab also switches to the chat's own files after the first message. #33954
- 🧭 Agents in unsupported model pickers. Connecting a terminal no longer adds its coding agents to Admin Models, base-model selectors, Arena candidates, task-model settings, group defaults, sub-agent settings, channel mentions or the Playground. Notes also stop choosing a terminal agent for title generation. Agents cannot be saved or imported as custom models or their base models, and their chat menu no longer offers Edit. Commit, Commit
- 🔌 Agent list after switching terminals. Changing terminals immediately removes the previous terminal's agents from the chat's model list while the new list loads. A model-list request from a chat you have left can no longer replace the current list. Commit
- 📋 Copy buttons over plain HTTP. At a plain HTTP address on your local network, copying comment links, search links, shared-chat links, webhook URLs, chat IDs, Git branch names, commit hashes and spreadsheet cells now works. Copy Formatted Text preserves formatting, and copied text keeps its final line break. #33939
- 📝 Clipboard prompts and Auto-Copy Response over plain HTTP. Prompts using the clipboard placeholder now insert into chats and channels even when the browser blocks clipboard access, with a message explaining that the clipboard could not be read. Auto-Copy Response stays off and shows a permission message instead of silently failing to turn on; turning it off no longer requires clipboard access. #33939
- 🧠 DeepSeek thinking mode with skills, knowledge and memories. These chats no longer fail with a missing reasoning error when the connection's Provider is set to "llama.cpp". DeepSeek thinking mode still requires that setting; connections using "Default" are unchanged. #33950
- 🏷️ Labels on grouped tool calls. With Tool Call Labels enabled, a group of tool calls keeps the labels that are present even if one call has none. The unlabeled call shows its tool name instead of making the whole group lose its labels. #33932
- 💬 Tool labels in later replies. The model now sees the labels it wrote for earlier tool calls when continuing a reply or answering another message, removing a reason it could stop writing labels as the conversation continued. #33953
- 🔎 Knowledge searches containing punctuation. Searches in knowledge files now find text such as "Art. 40" when the model marks the period as literal punctuation, instead of returning no matches. #33936
- 🪟 Side Chat with Temporary Chat disabled. Leaving Enforce Temporary Chat switched on while disabling Temporary Chat no longer prevents users from sending and saving side chats or dragging an existing chat into the side panel. #33930
- 🏠 Home folder tab title. A Files tab at the connected machine's home directory now says "Home" instead of showing the directory name, including when reopening saved tabs. Other folders keep their names. Commit
- 🦙 Missing Ollama in Docker images. Building the image that includes Ollama now fails if Ollama was not installed, instead of producing an image that cannot start Ollama after its download failed. #33944
- 📄 Errors after leaving a chat. Closing or leaving a chat while its tab layout is being saved no longer fills the server log with an error report for the dropped connection. #33929
Fixed
- 🔓 Chat over plain HTTP. Opening Open WebUI over plain HTTP at a LAN or Docker address now loads new and existing chats instead of staying on a spinner with "crypto.randomUUID is not a function". #33919
- 📝 Comments and Markdown editing over plain HTTP. Creating, replying to and editing comments, and editing Markdown files containing rendered blocks such as Mermaid diagrams or callouts, no longer fail with the same UUID error. Commit
- 🎛️ Side Chat permission switch. The Allow Side Chat switch is now visible even when Temporary Chat is off, so administrators can configure the two permissions independently. #33920
Added
- 🦾 Coding agents as models. On an Open Terminal that offers agents, coding agents such as Claude Code, Codex and Pi now appear in the model picker. You can chat with them or call them by voice, choose their reasoning effort and service tier, and approve their tool requests in the chat. Each conversation uses one agent, and agent chats cannot be regenerated, continued or edited and resent. If an agent's terminal goes away, its chats and new chats fall back to the default model and earlier replies keep the agent's name. Autocompletion in agent chats uses the task model, or is skipped when there is none. #33843, #33860, #33880, #33857
- 🪢 Introducing Tether. Tether is the new desktop companion app for macOS, Windows and Linux that ties your own computer to Open WebUI. Sign in once and your computer shows up under Your devices in the terminal menu, so models, coding agents and automations can work in your real files, repositories and tools, and you can reach them from any browser, including your phone. Tether connects out to Open WebUI, so there are no ports to open, its terminal output resumes after a dropped connection, and it delivers notifications to your desktop even when no Open WebUI page is open. It signs in with a renewable session that expires after 90 days without use by default, set with "SESSION_EXPIRES_IN".
- 💸 Much cheaper tool chats. Chats that use tools now cost much less. Citation instructions, memory recall, the skills list and attached knowledge are no longer rewritten into the system prompt or your message on every tool round, so the start of the conversation stays the same between requests and providers can serve it from their prompt cache. Custom pipes see a different message layout as a result; see Context as hidden tool results under Changed. #33852
- 🚦 Auto tool permissions. Administrators can now turn on Auto, a third tool permission mode next to Full access and Ask for approval (off by default). A review model checks each tool call and runs it right away only when it judges that no confirmation is needed. The review sends the tool, its arguments and your last few messages to the configured scoring API, and any call it cannot review, because of an error, a timeout or too much context, asks for approval.
- 📚 Full text for small files. Files whose extracted text is shorter than a new threshold (50,000 characters by default, set with "RAG_FULL_TEXT_CHAR_THRESHOLD" or in the admin Documents settings) are now included in full instead of through retrieval. This applies both when Preload files adds the content to the conversation and when the model reads an attached file with its file search tool, and the file viewer shows when it happens.
- ⏱️ Faster large chats. Large chats now open about twice as fast, and marking a chat as read or unread or removing a tag from it is 4 to 6 times faster. #33665, #33666
- 🏎️ Faster chat access checks. Sending messages, typing indicators, shared links, comments and file access no longer load the whole conversation just to check who may access it, which helps most on large chats. #33667, #33669
- 🔧 More forgiving tool calls. Tool calls with arguments wrapped in code fences, encoded twice or sent as text instead of numbers and booleans now run, and tool calls a model writes out as plain text markup are recognised and executed. #33868
- 🛑 Tool loop stops. A tool call that fails twice in a row with the same arguments now ends the tool loop so the model writes a final answer, and a reply cut off by the token limit no longer runs its unfinished tool calls.
- 🛠️ Default tools for models. The Model Defaults section in the admin Models settings now has a Tools picker, models without their own tool list start new chats with those tools enabled, and the new model form preselects them. #33749, #33787
- 🚀 Faster channels. The channel list, the channel messages and the replied-to messages in a channel now load much faster, especially with many channels or many replies, and new database indexes speed up channel threads and reactions. #33690, #33691, #33688
- 🪶 Lighter reasoning streams. Long reasoning replies from providers such as OpenRouter now send far less data to the browser and to Redis while they stream. #33698
- 🗂️ Faster sidebar chat list. The chat list in the sidebar now appears as soon as its own data is ready instead of waiting for the channel list. #33686
- ⌨️ Faster typing after large pastes. Typing in the chat input after pasting 16,000 lines of text is now about 117 times faster, down from 8.6 seconds to 74 milliseconds per keystroke. #33729, #12087
- 🧮 Faster Redis filter lookups. With Redis enabled, chat requests now look up the filters of a model faster. #33668
- 🪟 Split tab panes. On large screens, dragging a tab or a chat to the edge of the workspace, or pressing the new Split screen button, splits it into resizable panes, so several tabs, including the chat, stay visible side by side. Tabs can also be reordered and moved between panes with a long press and drag on touch screens. Commit
- 🧳 Tabs remembered per chat. The open tabs, split layout and selected terminal of each chat are now saved privately for each user on the server and restored when you reopen the chat, even on another device. Signing out warns you about unsaved file edits. #33899, #33885
- 🪣 Batched Tool Search calls. With Tool Search on, the model can run several independent tools it found in a single call_tool request, which saves model turns and so lowers the cost of tool-heavy chats.
- 🧭 Tool Search engines. Tool Search in the admin Interface settings can now match tools with embeddings from the Documents embedding model (kept in memory, not stored) or by sending the query and tool definitions to a relevance-scoring API, besides the default keyword search.
- 🏷️ Readable tool call labels. Administrators can now turn on Tool Call Labels in Admin Settings > Interface or with "ENABLE_TOOL_CALL_LABELS", off by default, to ask the model for short descriptions in the user's language of what each tool call is doing and what it did. Commit
- ✂️ Chat read limit. Admins can limit how much of an earlier chat the Chat History tools return, with "VIEW_CHAT_MAX_CHARS" or the new Chat Read Limit field in Admin Settings > General, and the newest messages are kept while older ones are cut. #33695, #33608
- 🌍 Client IP header placeholders. Custom headers of connections can now use "{{USER_IP}}" and "{{X_FORWARDED_FOR}}" to pass the client address on to the upstream provider, and both are empty for automations and background jobs. #33833, #33819
- ✉️ Non-ASCII email addresses. The sign-up page and the add and edit user dialogs now accept email addresses with non-ASCII characters. #33693, #33681
- 🕒 Reply finish time. Assistant replies now show the time they finished instead of the time the prompt was sent. #33895
- 🧐 Channel message search. Each channel header has a Search button that searches the messages of that channel.
- 🔔 Notifications open their chat. Clicking a desktop notification now brings Open WebUI to the front and opens the related chat, channel or calendar.
- 📮 MCP Apps message the chat. An MCP App can now send a text message to the chat, which you confirm first in a Confirm Prompt from Embed dialog. #33907
- 🖱️ Drag chats into side panel. Dragging a chat from the sidebar onto the chat area opens it as a side chat tab, and a side chat can now also be started from an empty new chat.
- 🪄 Quick open files. A new shortcut, Cmd+P (Ctrl+P elsewhere), opens a search box that finds files in the selected terminal by name or path and opens the one you pick in a tab.
- 🎹 Tab keyboard shortcuts. New shortcuts, which can be changed in Settings > Keyboard, open the Files, Desktop, Comments, new chat, new file, new terminal and new browser tabs, close and switch between tabs and toggle the split screen.
- 🗺️ Working directory picker. The message input has a working directory menu with folder search, a Recently Used list kept per user on the terminal and an Open Folder dialog.
- 📂 Folder default terminal. A folder can now have a default terminal and working directory in its Advanced settings, used for new chats in that folder.
- 🧾 Editable spreadsheets. Excel files in the file view can now be edited in the preview, with a formula bar, pasting from other spreadsheets and resizable columns, and saved back to the file, while formula cells and protected sheets stay read-only.
- 🖋️ Markdown and code editing. Markdown files in the file view now open in a rich text editor that saves back to Markdown, where pasted or dropped images are uploaded to the terminal, and the code editor suggests words that already appear in the file. Other text files and the Source view of previews now use the code editor directly, without a separate Edit step. Commit
- 🌳 Clone Git repositories. The file browser and the folder pickers can now clone a Git repository from an HTTP(S) or SSH URL into the chosen folder of the selected terminal.
- 🧬 Git changes in the editor. The code editor now marks the lines of a file that were added or changed since the last commit.
- 🔖 File browser Git status. Files and folders in the file browser are marked with a Git status letter such as M, A, D or U, and folders show that they contain changes.
- 🌿 Git panel feedback and resizing. The Git panel now shows what it is doing, such as staging, pushing or fetching, and its file list can be resized.
- 🧹 Discard any Git change. The Git panel can now discard staged and new files too, and its confirmation warns that files not in the last commit are deleted and that this cannot be undone.
- 🧲 Drag files into chat. Files from the terminal file browser can be dragged, one or several at once, into the message input to attach them, also with a long press on touch screens.
- 🗃️ Multiple Files tabs. The + menu can now open several Files tabs for the same terminal, each with its own folder and the folder name as its tab title, and moving files refreshes the other Files tabs on that terminal. Commit
- 🔲 Drag to select files. Several files in the file browser can now be selected by dragging a selection box over them.
- 🪂 Spring-loaded Files tabs. Holding a dragged file over a Files tab for half a second switches to that tab, so files can be moved into folders of another tab.
- ⚡ Faster terminal rendering. The terminal tab now renders with WebGL where the browser supports it and takes in bursts of output in larger batches, so heavy output is handled faster.
- 🩺 Terminal system info. The terminal menu now has a System info view with the terminal's CPU, memory, disk, network, uptime and top processes.
- 🖌️ Edit images from the terminal. The image edit tool can now take source images from the selected terminal by file path, as PNG, JPEG, WebP or GIF up to 25 megapixels. #33824
- 🎙️ Voice notes in files. A new shortcut, Cmd+Alt+M (Ctrl+Alt+M elsewhere), records a voice note, can transcribe it, and saves the recording with a Markdown note in the current folder of the selected terminal, then opens the note in a tab. Escape cancels a recording. #33902, #33903
- 📢 Model availability events. Open WebUI now publishes model.available and model.unavailable events when a model appears in or disappears from the fetched model list.
- 🗒️ Descriptions in model pickers. The tool, skill, filter, action and sub-agent pickers in the model editor now show each item's description under its name. #33723
- 🔢 Grouped token counts. Whole numbers in the token usage tooltip are now grouped by thousands for the interface language, for example 26,678,623. #33896
- 🎼 Notification sound styles. The notification sound setting now offers Chime, Soft, Bell and Tap sounds, with a play button to preview each one.
- 🔄 General improvements. Various improvements were implemented across the application to enhance performance, stability, and security.
- 🌐 Translation updates. Translations for Portuguese (Brazil), Azerbaijani, Basque, Bengali, Bosnian, Bulgarian, Cebuano, Croatian, Czech, Danish, Dutch, Estonian, Faroese, Filipino, Galician, Georgian, German, Greek, Hebrew, Hindi, Hungarian, Indonesian, Irish, Japanese, Kabyle, Korean, Lithuanian, Malay, Norwegian Bokmål, Persian, Polish, Punjabi, Romanian, Russian, Serbian, Slovak, Slovenian, Spanish (Spain), Swedish, Tamil, Thai, Tibetan, Turkish, Turkmen, Ukrainian, Urdu, Uyghur and Vietnamese were enhanced and expanded.
Fixed
- 🛡️ Security Advisory: This release contains security fixes. We strongly recommend updating all deployments as soon as possible to have the latest security fixes in place. These fixes are not listed in this changelog; we will publish the security advisories for them immediately following CVE assignment. Advisories
- 🗄️ Redis Cluster chats. Sending a chat message on a Redis Cluster deployment no longer fails while the personal terminal relay starts. #33766
- 🚪 Automatic SSO sign-in. With automatic SSO redirect on, an expired session now signs you back in without a state mismatch error. #33791, #33783
- 🎨 OpenAI image default model. Image generation with the OpenAI engine now defaults to GPT-IMAGE 2 instead of the shut-down DALL·E 2, and the shut-down DALL·E models are no longer offered. #33694, #33578
- 🖼️ Default GPT image requests. OpenAI image generation and editing with no model explicitly set no longer send an unsupported response format to the default GPT image model. #33875
- 🔐 Terminal Session authentication. Terminal connections set to Session authentication now accept an API key, so Verify Connection, the policy settings and the terminal tools work again. #33699, #30001, #33579
- 🦙 Ollama broken tool calls. Chats with Ollama models no longer get stuck after a tool call with cut-off, rejected or empty arguments. #33697, #33671
- 🧠 Complete compaction summaries. Context compaction summaries are no longer cut off at 1000 tokens, which could leave reasoning models with an empty or truncated summary. #33726
- 💾 Prompt cache on tool results. Tool results sent through OpenAI-compatible gateways now keep their prompt cache marker, so Anthropic models behind OpenRouter or LiteLLM can cache the prefix in tool loops. #33757, #33755
- 📖 File read size limit. Reading a file by line range now stops at the same size limit as other reads, so one read can no longer overflow the context window of the model. #33696, #33607
- 🧰 Tool Search retries. A malformed tool call with Tool Search on no longer disables tools for the rest of the reply, so the model can correct its arguments and try again. #33872
- 🧩 Split tool calls. When a model sends several tool calls glued into one, the chat no longer shows an extra tool call that never gets a result. #33779, #33778
- 🔎 External knowledge base queries. External knowledge bases now search with every generated query at the same time instead of using only the results of the first one. #33704, #33184
- 📊 Large spreadsheet uploads. Uploading a large, sparsely filled spreadsheet no longer takes minutes and stalls the server. #33770, #33768
- 🧱 Milvus knowledge base uploads. Adding a file to a knowledge base on Milvus no longer processes the file a second time or stores its chunks twice. #33745, #33744
- 📄 MinerU 4.1 servers. The MinerU document loader in local mode now works with MinerU 4.1 servers, where every document used to fail. #33719, #33718
- 🔑 MCP OAuth stays signed in. The login to an MCP tool server that uses OAuth is no longer lost when the provider has a temporary outage while the token is refreshed. #33805, #33803
- 📍 Location lookup delays. With location access turned on, sending a message no longer waits for a slow or unanswered browser location request.
- 🗓️ Automation schedule examples. Models creating automations through the built-in tool no longer hit "Invalid RRULE" errors caused by a wrong example schedule format. #33732, #33730
- ♿ Keyboard and screen reader fixes. Keyboard focus now stays where it belongs: on buttons while they save, in place when you cancel a message edit, switch message versions or sort and page a workspace or notes list, on the first and last page buttons, and in the title field while a note title is generated. Closing a channel thread, a note panel or a chat menu returns focus to where it was, and the sign-in skip link moves focus to the form. The show password button is now wide enough to hit, and holding Shift no longer swaps row buttons for Delete while a keyboard user has focus.
- 🕵️ Audit log secret masking. With request auditing on, the audit log now masks API keys, tokens and secrets saved in the admin settings, not only passwords. #33746
- 🔇 Queries kept out of logs. The default server log no longer contains your search queries or the provider address of a failing external reranker. #33701
- 💬 Chat message search. Workspace search now finds message text and shows matching excerpts in chats whose messages are stored separately from the chat record. Commit
- 🔃 Search results stay current. Notes, archived and shared chats, the admin user list and the knowledge and file pickers in the message input no longer let an older search replace the results of a newer one. Commit
- 👎 Feedback details saved. A thumbs-down reason and comment are no longer lost when saved before the feedback tags finish generating. #33716
- 🎒 Side chat default tools. New side chats now start with the same default tools, skills, filters and web search as a normal chat with that model. #33788
- ✅ Side chat tool approvals. Tool calls in side chats now show working approval controls and resume the reply after approval. #33889
- ❓ Side chat questions. Questions asked through the model's ask_user tool now appear in side chats and can be answered or dismissed, including after reopening the chat. #33901
- ⏩ Continued replies on Responses connections. Continuing a reply on a Responses API connection now adds the new text to the existing reply.
- 🛂 Model required before sending. The file ask box, follow-up suggestions and queued messages now report "Model not selected" instead of trying to send when no model is selected. #33838
- 👻 Temporary chat button with tabs. The Temporary Chat button is back on new chats when the tabbed layout is on. #33680
- 🪧 Stable MCP app views. MCP apps in chat no longer reload and lose their state while the message around them updates.
- 🎴 Artifacts keep the chat visible. Opening an HTML or SVG preview from a reply now uses the side pane instead of replacing the visible chat tab. #33894
- 📣 Banners in tabbed view. Admin banners and license notices now show on the new chat page when the terminal tabs split the screen. #33789
- 📡 Tabs over plain HTTP. The New File, New Terminal and Browser tabs in the side panel now open when Open WebUI is reached over plain HTTP on a LAN or Docker address. #33847, #33845
- 🕶️ Per-chat terminals in temporary chats. Per-chat workspace terminals are no longer offered in temporary chats, where sending to them failed, and one already selected is deselected when Temporary mode is turned on. #33866, #33861
- 🖥️ Terminal kept on model switch. Switching to a model without its own default terminal no longer turns off the terminal selected in the chat.
- 📁 Working folder in new chats. The terminal working folder of a new chat is no longer reset to the root when the chat is created before the file browser has finished loading.
- 📞 Return to call. The Return to call button now reopens the active voice call after switching to the chat tab. #33898
- 🔕 Channel notifications clear on read. Opening a channel now marks its notifications as read.
- 🪝 Webhooks window stability. The channel Webhooks window no longer jumps back to the webhook list when a new message arrives, so edits in progress are kept. #33776, #33775
- 🎧 Background audio on iOS. With Auto-Playback Response off, tapping in Open WebUI or opening a chat no longer stops audio playing in other apps on iOS. #33780
- 🖨️ Scanned PDF preview. PDF previews no longer show blank pages for scanned PDFs that store their pages as JPEG2000 images. #33906, #33905
- 🔬 Preview zoom stability. Zooming in PDF, Word and PowerPoint previews no longer resets or drifts when you switch tabs, change slides or resize the pane. #33836, #33835
- 📌 File preview from Controls. A file preview opened from the Controls menu no longer closes when you click inside it. #33831, #33828
- 🔭 Tab bar and UI scale. Tab content no longer slides under the tab bar when the UI scale is above 1. #33830, #33829
- 🔘 Desktop app top clicks. In the desktop app, banners and menus near the top of the window now take clicks instead of dragging the window.
- 📱 Mobile menu keyboard. Closing the + menu on a phone no longer opens the keyboard. #33893
- 📝 File rename in terminal. Renaming a file or folder in the Open Terminal file browser no longer shows a false error after the rename succeeded. #33786
- 📋 Stable file selection. Selecting files no longer shifts the file rows when the bulk action bar appears, and clicking an empty area of the file list clears the selection. Commit
- 🫥 Selection popups on scroll. Scrolling the file view dismisses its Ask box, and scrolling a rich text editor dismisses its floating formatting menus until the selection changes. Commit, Commit
- 🙈 Empty control groups hidden. When every control of a group is hidden with Visible controls, the empty group heading or submenu no longer shows. #33867
- 🌄 Chat background behind tabs. The chat tab bar no longer covers the chat background with a solid strip. #33849
- 🌫️ Top bar blur. The top bars of Workspace, Admin Panel, Playground and Calendar no longer blur a page background image. #33799, #33798
- 🗑️ Feedback model filter. Deleting the last feedback of a model in the Evaluations feedback list now clears the model filter. #33794, #33793
- 🔤 Visible controls labels. Six toggles under Visible controls in the Admin Interface settings now show their names instead of raw translation keys. #33734
- 🤫 Quiet terminal disconnects. Closing the file panel or a terminal view right after opening it no longer logs an error traceback on the server. #33863, #33850
- 📜 Quieter MCP connection logs. Logs no longer fill with a warning for every MCP server a user has not connected yet each time they load Open WebUI. #33663
- 🪪 OAuth token refresh warnings. Refreshing OAuth tokens with a client secret no longer logs deprecation warnings on every call. #33700, #33297
Changed
- ⚠️ Database Migrations: This release includes database schema changes; we strongly recommend backing up your database and all associated data before upgrading in production environments. If you are running a multi-worker, multi-server, or load-balanced deployment, all instances must be updated simultaneously, rolling updates are not supported and will cause application failures due to schema incompatibility.
- 💥 Preload files off by default. With native function calling, the default, attached files are no longer added to the conversation automatically, and models read them with their file tools instead. Models without tool support or without the Files built-in tool can no longer see attached files until you turn on Preload files in the model's Capabilities. Legacy function calling, and API requests that do not come from a chat in the web UI, keep the old behavior. #33815
- 🎯 Retrieval defaults. Unless you saved your own values, "RAG_TOP_K" and "RAG_TOP_K_RERANKER" now default to 10 instead of 3, and "CHUNK_SIZE" to 1024 instead of 1000. Each query now retrieves more text, so prompts with knowledge get longer. The new chunk size applies to files processed from now on; existing knowledge keeps its chunks until it is reindexed.
- 🧷 Context as hidden tool results. With native function calling, memory recall, the skills list and the attached knowledge list now arrive as hidden tool results after your message instead of in the system prompt. This is what keeps tool chats cacheable, and it means custom pipes can see a tool message as the last message.
- 🗨️ Side chats are saved. Side chats, which were always temporary in 0.13.0, are now saved as regular chats and appear in the chat list. They stay temporary only when you turn on Temporary Chat in the side chat or your admin enforces temporary chats.
- 🚧 Tool call limit. Reaching the tool call limit now makes the model write a final answer without tools, instead of ending the reply with the "Tool-call limit reached" error.
- 🔦 Tool Search prompt. With Tool Search on, the model no longer sees a list of the hidden tools and finds them only by searching.
- ✏️ Always Use Full Text. The Full Context Mode setting in the admin Documents settings is now called Always Use Full Text and includes complete documents regardless of the full text threshold.
- ↔️ Always full-width tabs. On large screens the tabbed workspace now always fills the chat area, with the chat as one of its tabs, and the full-width toggle added in 0.13.0 is gone.
- 🔀 Integrations moved into + menu. The Integrations button is now part of the + menu in the message input, and the new Show Input Selections setting, off by default, controls whether selected tools, skills, filters and capabilities are shown beside it.
- 🔓 Side chat permission. The Allow Side Chat permission no longer requires the Temporary Chat permission.
- 🕸️ Chrome first in Browser tab. The Browser tab now starts in Chrome mode when the terminal offers it, and switches to Proxy mode with a message if Chrome fails to start.
- ⛔ Terminal select setting removed. The Show Files on Terminal Select setting has been removed from the Interface settings. #33855, #33840
AI summary Added tabbed workspace, desktop, and browser tabs
Added
- 📑 Tabbed workspace. The Tabs button at the top of a chat now opens a workspace for files, multiple terminal sessions, artifacts, previews, voice calls and comments. Add tabs from the + menu, drag to reorder them, and create and save new files directly into the selected terminal. Keep the conversation and its working files together while coding, analyzing data or reviewing generated documents. Commit, Commit, Commit, Commit
- 🖥️ Open Terminal Desktop tab for computer use. Open the Desktop tab from the + menu to view and control the desktop provided by your connected Open Terminal, with mouse, keyboard, clipboard and audio support where available. With a model that can call its desktop tools, you can follow computer-use workflows in the same desktop and step in yourself: operate graphical apps, inspect generated documents or handle a step that needs manual input. Requires an Open Terminal connection with desktop support enabled. Commit, Commit, Commit, Commit
- 🌐 Browser tab for browser use and web app testing. Open websites and local development servers beside your chat, with navigation controls, multiple tabs and a mobile viewport. On a compatible Open Terminal connection, Chrome mode gives you and the model a shared browser session for browser automation: navigate sites, interact with forms and test the web apps the model builds. You can watch the work and interact with the page yourself; Proxy mode remains available for previews. Requires browser support on the selected terminal, and Chrome availability for browser-use tools. Commit, Commit, Commit
- ↔️ Full-width tabs. On larger screens, expand the tabbed workspace to fill the chat area, with the main conversation kept as its own tab. Switch between the chat, code, browser and desktop with more room to work, or return to the side-by-side layout; your layout choice is remembered. Commit, Commit
- 💻 Open WebUI Desktop 0.1.x. The separately installed app for macOS, Windows and Linux now focuses on setting up and running Open WebUI on your own computer, with optional llama.cpp local models. The 0.1.0–0.1.4 updates bring a slimmer interface, searchable settings, startup logs, optional automatic Desktop updates, start-to-tray controls and image copying and saving. They also improve screen sharing, knowledge folder uploads, Linux graphics and shortcuts, and NVIDIA/AMD local-model setup and updates. Use it for a managed local installation and local AI workflows. Desktop changelog, Download v0.1.4
- 🌿 Git in the side panel. When the selected terminal's folder is a Git repository, a Git panel in the tabbed side panel now lets you review diffs, stage, discard and commit changes, manage branches and worktrees, push and pull, browse history and work with GitHub pull requests, and open files can show their changes with Show diff. Commit, Commit, Commit, Commit, Commit
- ⚡ Live terminal updates. On terminals that support it, the Git panel, the list of open ports and the running processes and their output now update as soon as something changes instead of every few seconds. Commit, Commit
- 💭 Side chats. Typing "/side" or "/btw" in a chat, or choosing Side chat in the side panel, now opens a temporary side conversation using the chat so far as context. Ask a quick question or explore an alternative without adding the detour to the main conversation; the side chat disappears when you close its tab. Available to administrators and users given the new Allow Side Chat permission, which is off by default. Commit, Commit, Commit
- 💬 Comment threads. Administrators can now turn on Comments in Admin Settings > General or with "ENABLE_COMMENTS", which lets users leave comments with replies, reactions and mentions on the chats, notes, models, knowledge bases, prompts, skills and tools they can open, including on a selected passage of a chat message, which then stays highlighted in the reply, or of a note page, shown in a Comments tab in the chat's side panel, which "/comments" opens, and with comment counts in the workspace lists. Commit, Commit, Commit, Commit
- 🔔 Notification inbox. A new Inbox in the sidebar collects comments on your items, replies to your comments, mentions in comments and channels, replies in channels, failed automation runs and upcoming calendar events with an unread count, and administrators can turn it off in Admin Settings > General or with "ENABLE_NOTIFICATIONS". Commit, Commit, #19539
- 🧩 MCP Apps in chat. Tools from MCP servers that come with an interactive app now show that app right in the chat after they run, in a sandboxed frame that can only reach the sites the server declares, and every tool call the app wants to make waits for you to confirm it. Commit, Commit, #21040, #20915
- ✨ Generate title from top bar. The rename field at the top of an open chat now has the same sparkle button as the sidebar, which generates a title from the conversation and saves it right away. #33382
- 📐 Side panels on tablets. On screens from 768 pixels wide, such as tablets and narrow windows, channel threads and the pages list and chat in notes now open beside the content instead of over it, picking a terminal opens its files beside the chat as well, and the search dialog shows its preview there by default and remembers whether you turned it on or off at any window size. Commit, Commit
- 📊 More PowerPoint charts. The PowerPoint preview now draws pie, doughnut and horizontal bar charts as well as values below zero, and shows "Chart preview unavailable" for charts it cannot draw. Commit, #32884
- 🧊 3D model previews. GLB, glTF, STL, OBJ and PLY files in a terminal's file browser and in files shown in chat replies now open in a 3D viewer that you can rotate and zoom. Commit
- 🔘 Switch tools on and off. Tools in Workspace > Tools now have an on and off switch for everyone who can edit them, and a tool that is switched off can no longer be picked or run in any chat, including chats where it was already selected. Commit
- 🛠️ Tool creation from chat. Models with native function calling can now create a private Python tool in your workspace when you ask them to, for administrators and users allowed to create workspace tools, and it can be turned off per model under Tool Creation in the model editor's builtin tools. Commit
- 🤝 Multi-model sub-agents. The model editor now lets you choose other models for a model to delegate work to, enabling multi-agent workflows such as assigning research, coding and review to different models. The parent model chooses among the configured models the user can access; previously, sub-agents always used the same model as the parent. Commit
- 📏 Knowledge base file limits. Administrators can now cap how many files a knowledge base holds, how large each of its files may be and how many files go into it in one upload, for all knowledge bases in Admin Settings > Documents or with "KNOWLEDGE_MAX_FILE_COUNT", "KNOWLEDGE_MAX_FILE_SIZE" and "KNOWLEDGE_MAX_UPLOAD_COUNT", and a knowledge base's owner can set its own file count and size limits under Advanced when creating it or from Settings in its menu, while the global limits always apply. Commit, Commit, #33318, #24845
- 📄 File sizes in previews. Opened terminal files now show their size in the file toolbar, so you can check a document or generated output without leaving its preview. Commit
- 🗂️ Sort files by date added. The terminal file browser can now sort files by when they were added, next to name, size and date modified, on terminals that report creation times. Commit
- ↕️ Move blocks in notes. Every paragraph, heading and other block in a note now has a handle beside it that you can drag to move the block, or click to move it up or down. Commit
- 📋 Copy code in notes. Code blocks in notes now have a Copy button and use clearer syntax colors in light mode. Commit, Commit, #33455
- ⏸️ Approvals shown in sidebar. Chats in the sidebar now show a "Waiting for approval" marker while a tool call in them waits for you to allow or deny it. Commit, Commit
- 🏷️ Role prompt variable. System prompts, saved prompts and chat requests through the API can now use "{{USER_ROLE}}", which is replaced with the signed-in user's role, such as admin or user. #33441, #4765
- ♿ Accessibility improvements. Screen reader users now hear a name for every window, form field, setting and icon button, including Admin Settings, Valves, filters and channel message actions, and keyboard users can answer tool approvals, dismiss notifications, open model details in Admin Analytics, sort Analytics and group member tables, resize the sidebar, open notebook cells for editing, use multi-select lists in Valves, pick a status emoji, choose tag and voice suggestions, open the sources of a web search, use the menus in the Search window, stay inside the full-size image viewer and side drawers until they close them and reach every event on the calendar. Closing a window opened from a menu now returns focus to the menu button, Tab cancels shortcut recording, task-list checkboxes are named by their text, and unselected Analytics tabs have better contrast in Accessibility Mode. Commit, Commit, Commit, Commit
- 🔄 General improvements. Various improvements were implemented across the application to enhance performance, stability, and security.
- 🌐 Translation updates. Translations for Portuguese (Brazil), Latvian, Azerbaijani, Basque, Bengali, Bosnian, Bulgarian, Cebuano, Croatian, Czech, Danish, Dutch, Estonian, Faroese, Filipino, Galician, German, Greek, Hebrew, Hungarian, Indonesian, Irish, Japanese, Korean, Lithuanian, Malay, Norwegian Bokmål, Persian, Polish, Punjabi, Romanian, Russian, Serbian, Slovak, Slovenian, Spanish (Spain), Tamil, Turkish, Turkmen, Ukrainian and Vietnamese were enhanced and expanded.
Fixed
- 🛡️ Security Advisory: This release contains security fixes. We strongly recommend updating all deployments as soon as possible to have the latest security fixes in place. These fixes are not listed in this changelog; we will publish the security advisories for them immediately following CVE assignment. Advisories
- 🔃 New chats stay in sync. A chat started from the home page keeps syncing with the server again after it is created, so files removed from the chat's Controls no longer come back after a reload and the owner of a chat shared with Allow replies gets Regenerate on other people's replies again. #33377, #33376
- 📂 Temporary Chat on folder pages. Turning Temporary Chat on or off on a folder's page, with the button at the top or with "/temporary", now keeps you on the folder page, so the next chat is still saved in the folder and gets its system prompt, default model and knowledge. #33456, #33448
- 🧵 Thread reply focus. Choosing Reply in Thread on a channel message now puts the cursor in the thread's reply box, also when the thread panel is already open. #33467, #33466
- 🪟 Thread message actions visible. Hovering the top message in a channel thread now shows its whole actions toolbar instead of cutting off its top half at the edge of the panel. Commit, #33472
- 📞 Interrupting realtime calls. Interrupting the model while it speaks in a realtime voice call no longer ends the whole call when the reported playback position does not match the audio. Commit, #33550
- 📶 Temporary chats after reconnecting. A reply that is still being written in a temporary chat now keeps arriving after the connection drops and comes back, and a reply that could not finish shows "Connection lost. Please retry." instead of loading forever. Commit
- 🏠 New chat initialization. Opening or reloading the home page or a folder now selects the model and applies chat URL parameters again, instead of leaving the model empty and ignoring the link’s settings. Commit
- ✍️ Restored chat drafts. Saved draft text now returns to the message input after the new-chat editor mounts, instead of disappearing during initialization. Commit
- 🧩 Nested code previews. Preview now opens code blocks inside lists, quotes and details blocks, so embedded examples and generated artifacts can be inspected directly. Commit
- 🖼️ Image drops from photo apps. Images dropped into chats and channels are now recognized from their contents when the source omits the file type, including photos dragged from macOS Photos, instead of being treated as generic files. Commit
- 📏 Live context usage estimates. The context indicator now accounts for output arriving after the latest reported token usage and avoids counting assistant text twice when it also appears in structured output, making it easier to track context-window usage during long agent runs. Commit
- 🔌 Responses-backed API streaming. Clients using Open WebUI’s Chat Completions API now receive the expected streaming format even when the upstream model uses the Responses API, including text, reasoning, tool calls, token usage and completion reasons. Commit
- 😀 Channel reaction names. Hovering a reaction on a channel message now shows who reacted when those names are available, instead of showing only the count. Commit
- 📊 Analytics activity scrolling. The activity chart now tracks the element actually being scrolled, keeping its visible rows in sync when browsing longer activity lists. Commit
- 🔑 Saved OAuth provider settings. OAuth providers now load their saved connection settings when Open WebUI starts. Admin Settings also explains that changes to provider connections require restarting every Open WebUI instance. Commit
Changed
- ⚠️ Database Migrations: This release includes database schema changes; we strongly recommend backing up your database and all associated data before upgrading in production environments. If you are running a multi-worker, multi-server, or load-balanced deployment, all instances must be updated simultaneously, rolling updates are not supported and will cause application failures due to schema incompatibility.
- 🎛️ Controls in the model menu. The chat files, valves, system prompt and advanced parameters moved from the Controls panel at the top of the chat to Controls in the model controls menu next to the message input, and the button at the top now opens the tabbed side panel. Commit, Commit
- 🗺️ Overview in the chat menu. The chat Overview, the map of all messages and their branches, now opens as a full-screen window from Overview in the chat's menu instead of as a tab in Chat Controls. Commit
- 🔁 Higher tool-call limit. A single response with native function calling can now run up to 1024 rounds of tool calls before it stops with "Tool-call limit reached", up from 256, and setting "CHAT_RESPONSE_MAX_TOOL_CALL_ITERATIONS" to 256 keeps the old limit. #33384
- 🔓 No sub-agent limits by default. The sub-agent limits in Admin Settings and through "SUBAGENTS_MAX_CONCURRENT", "SUBAGENTS_MAX_ASYNC", "SUBAGENTS_MAX_ITERATIONS" and "SUBAGENTS_MAX_OUTPUT" now default to no limit, and leaving one empty means no limit, where they defaulted to 20 at once, 20 in the background, 30 tool loops each and 30000 characters of output. Commit
- 🧠 Context compaction and prompt caching. When the same model handles the conversation and its compaction summary, eligible requests now reuse the conversation’s prompt structure so the provider can use its prompt cache. This can reduce the cost of summarizing long conversations and agent workflows; other requests retain the separate summary path. Commit
- 💻 Desktop 0.1.x upgrade changes. Open WebUI Desktop no longer switches between remote servers, bundles Open Terminal, or provides the old Spotlight and global voice overlays. Existing remote-connection users should review Open WebUI Tether. The Open Terminal Desktop and Browser tabs described above use a separately configured Open Terminal connection. Desktop 0.1.0 release notes
- 📭 File processing alerts removed. Personal notifications in Settings > Notifications no longer offer "Retrieval Content process failed", so notification webhooks no longer receive an alert when a file fails to process. Commit
Added
- 🧭 Analytics session explorer. Administrators can now open a Sessions tab in Admin Analytics that lists each chat with its tokens, responses, errors and time span, filters them by search, model, token range or sub-agents, and shows a chat's saved replies and tool calls with their timing and outcome, as long as "ENABLE_ADMIN_CHAT_ACCESS" is on. Commit
- 📑 Pages in notes. Notes can now hold several pages and subpages, which you can add, rename, reorder, move under another page, download, link to and delete from the page tabs, and search results and the model now work with notes page by page. Commit
- 👀 Live presence in notes. A note now shows who else has it open, and clicking a person jumps to the page and spot they are working on. Commit
- ✏️ Rename from chat menu. The menu next to an open chat's title now has Rename, which turns the title at the top of the page into a text box, so a chat can be renamed without finding it in the sidebar first. #33177, #33176, #33246, #33339, #33338, Commit, Commit
- 🖼️ Generated images in terminals. When a terminal is selected in a chat, generated and edited images are now also saved to its working folder, so the model can keep working on them with the terminal's tools. Commit
- 🗃️ Models show chat files. Models can now show a file attached in the chat or stored in an attached knowledge base right in the reply with the built-in display_file tool, which until now only worked for terminal files. #32471, Commit, Commit
- 💾 Chat model choice saved. Changing the model in a chat is now saved with the chat right away, so the chat reopens with that model even when no message was sent after the change. Commit
- ⌨️ Unassign single shortcuts. Settings > Keyboard now shows a small x beside every assigned shortcut that unassigns only that one, so a shortcut such as Regenerate Response on Ctrl+R or Cmd+R can be switched off while all others keep working. #33185, #31463, Commit
- 🎚️ Slider for number Valves. Plugin authors can now show a number Valve or UserValve as a slider with a number box beside it, whose ends come from the field's minimum and maximum, so users see the allowed range while picking a value. #33179, #32045, Commit
- 📇 Provider details for models. Administrators editing a model now see a Provider details section with the input and output types, context window, maximum output and supported parameters that the provider reports for it. Commit
- 🧲 Embedding prefixes in settings. Administrators can now set the query and content prefixes that some embedding models need under Admin Settings > Documents, where they could only be set with the "RAG_EMBEDDING_QUERY_PREFIX" and "RAG_EMBEDDING_CONTENT_PREFIX" environment variables. Commit
- ⏱️ Terminal start timeout. Open WebUI now waits up to 60 seconds by default when it asks a terminal for its skills, AGENTS.md file or uploads, so a terminal that is still starting can answer, with "AIOHTTP_CLIENT_TIMEOUT_TERMINAL" to change the limit and a message naming the terminal when it still times out. Commit
- 🚨 Automation failure notifications. Automation run failed is now an event users can pick for their notification targets under Settings > Notifications, so the owner hears about every failed run with the automation's name and error. #33331, #33228, Commit
- ⚡ Faster knowledge base commands. Knowledge base commands such as ls, cat and grep and the knowledge base sync check now list files without loading every file's full text, so on large knowledge bases each call is up to ten times faster and uses far less memory. #33332, #33322, Commit
- 🔭 Faster chat search. Searching your chats, including from the Reference Chats menu and the model's chat search tool, no longer loads every matching chat in full, so results come back faster on large chat histories, and a result's message preview in the Reference Chats menu now loads when you hover it. Commit
- 📦 Batched document embedding. Large documents are now embedded and saved in batches of 1000 chunks by default, set with "RAG_EMBEDDING_BUFFER_SIZE", so processing a very large file holds far less in memory at once. Commit
- 🧾 Mistral OCR model choice. Administrators can now pick the Mistral OCR model and send extra request options through a Parameters field in Admin Settings > Documents or "MISTRAL_OCR_PARAMS", for example to stay on an older OCR model or to match what an Azure AI Foundry endpoint expects. #33333, #19707, Commit
- 📉 Negative relevance threshold. The Relevance Threshold in Admin Settings > Documents now takes negative values for rerankers that return raw scores below zero, where the field refused them and a negative "RAG_RELEVANCE_THRESHOLD" kept the page from saving while hybrid search was on. #33330, #29888, Commit
- 🎞️ Animated ComfyUI images. Images from ComfyUI's Save Animated WEBP and Save Animated PNG nodes now show up in the chat as the animated file, for both image generation and image editing. #33325, #33013, Commit
- 📡 Metrics skip health checks. OpenTelemetry metrics now leave out requests whose URL matches "OTEL_PYTHON_FASTAPI_EXCLUDED_URLS" or "OTEL_PYTHON_EXCLUDED_URLS", so health checks and probes no longer fill the request metrics, the same way traces already skip them. #33329, #31609, Commit
- 🗝️ Admins manage user API keys. Administrators can now rotate or delete a user's API key in the Edit User dialog, and a rotated key is shown once so it can be copied. #32774, Commit, Commit
- 👓 More hideable chat controls. Administrators can now also hide the selected tools, skills, filters, web search, image generation and code interpreter shown in the chat input, through Visible controls in Admin Settings > Interface or "UI_VISIBILITY". #32408, Commit
- ♿ Improved UI accessibility. Screen reader and keyboard users can now move through a chat message by message with each message's name as a heading, move through menus and submenus with the arrow keys, Home and End, reach a note's Generate title button and the options of a filter list opened with the mouse, leave the Compare revision editors with Escape and Tab, and hear names for every field of the calendar event window, the last column of the Users and Feedback tables and each day in the Settings > Usage heatmap, while the model picker outlines only one row, takes keyboard focus when opened in a window and leaves focus in that window when it closes, and focus outlines on tab rows are no longer cut off.
- 🔄 General improvements. Various improvements were implemented across the application to enhance performance, stability, and security.
- 🌐 Translation updates. Translations for Portuguese (Brazil), Catalan, Italian, Azerbaijani, Basque, Bengali, Bosnian, Bulgarian, Cebuano, Croatian, Czech, Danish, Dutch, Estonian, Faroese, Filipino, Galician, Georgian, German, Greek, Hebrew, Hindi, Hungarian, Indonesian, Irish, Japanese, Kabyle, Korean, Latvian, Lithuanian, Malay, Norwegian Bokmål, Persian, Polish, Punjabi, Romanian, Russian, Serbian, Slovak, Slovenian, Spanish (Spain), Swedish, Tamil, Thai, Tibetan, Turkish, Turkmen, Ukrainian, Urdu, Uyghur and Vietnamese were enhanced and expanded.
Fixed
- 🛡️ Security Advisory: This release contains security fixes. We strongly recommend updating all deployments as soon as possible to have the latest security fixes in place. These fixes are not listed in this changelog; we will publish the security advisories for them immediately following CVE assignment. Advisories
- 🔎 Search dialog filters restored. The search dialog again understands and suggests the "tag:", "folder:", "pinned:", "shared:" and "archived:" filters, where adding one emptied the results and archived chats could not be found at all. #33181, #32518, Commit
- 📍 Search locations by type. The Location filter in the search window now lists only folders for Chats and only channels for Messages, and switching the content type clears a location that no longer fits, where such a pick returned an empty list. #33183, #33182, Commit
- 🧮 Inflated streamed token counts. Token counts for a reply are no longer inflated when the provider reports its usage again on every streamed chunk, where every report was added up, so the usage shown for the reply and in Admin Analytics now matches the provider's final count. #32052, #29064, #28528, Commit
- 📝 Note typing after Formatting. Typing in a note now works right after switching Formatting on or off in the note's menu, where the editor stopped taking input and saving until the page was reloaded. #33222, Commit
- 🎙️ Note recordings kept. A recording in a note now stays in the note when the connection drops while it is being transcribed, for example when a laptop goes to sleep. #33240, Commit
- 🀄 IME typing with Backspace. Pressing Backspace while composing text with a Japanese or other input method in the chat input no longer ends the composition or makes the text appear twice, which started in 0.12.4. #33245, #33223, Commit
- 🤖 Photos for sub-agents. Models can now hand a photo attached in the chat to a sub-agent, where the hand-off failed with "file_ids not attached or unavailable" and no sub-agent started. #33244, #33241, Commit
- 🪆 Sub-agent partial output kept. A foreground sub-agent that stops at the tool-call limit now hands the model its partial output along with the error, where everything it had written was lost, and the error now reads as a plain message in both foreground and background. #33269, #33267, Commit
- 🔗 MCP failures keep the chat. A tool call to an MCP server that cannot be reached, for example after a restart or a dropped connection, now fails on its own and the reply still finishes, where the whole chat stopped as if you had pressed stop. #33272, #33268, Commit
- 🔃 Cut-off replies stay finished. A reply cut off by a server restart no longer goes back to the loading dots when the connection comes back later. #33277, Commit
- 🐢 Slow chat loads. Opening a chat on a slow connection or switching chats quickly now shows the chat you opened, where a slower earlier load could replace it with an old copy and the next message forked the chat. #33180, Commit
- 👁️ Tool images with Vision off. Models with Vision turned off now get a short "[image]" note in place of each image a tool returns, where the reply failed with an error and every later message in that chat failed the same way. #33345, #32720, Commit
- ✋ Stop button on others' replies. In a chat shared with Allow replies, the Stop button no longer shows on someone else's running reply, where it showed but did nothing. #33343, Commit
- 🧷 Refused sharing changes undone. The Access dialog of a prompt, skill, tool, note or knowledge base now goes back to the saved sharing when the server refuses a change, where it kept showing the refused change as if it were saved, and the model editor now shows the server's message for such a refusal. #33209, #33208, Commit
- 🔌 Tool server Headers reset. Adding another connection in External Tool Servers now starts with empty Headers, where it kept the previous connection's Headers and saved them on the new one. #33201, #33200, Commit
- 🧰 OpenAPI tools listed. The Available Tools window in a chat now lists the tools of an OpenAPI tool server, where it said "No tools found". #33205, #33204, Commit
- 🎛️ Generation switches update. Saving Title, Follow-Up, Tags or Autocomplete Generation in Admin Settings > Interface now updates the matching switches in Settings > Interface and autocomplete in the chat input right away, where they kept the old setting until a reload. #33197, #33196, Commit
- 🗂️ Local knowledge filter. Picking Local in the source filter on Workspace > Knowledge now lists knowledge bases made with Create, where it showed "No knowledge found". #33199, #33198, Commit
- 🪣 Large S3 Vectors deletes. Removing a file or knowledge base with more than 500 chunks from S3 Vectors now works, where the delete request went over the service's limit of 500 keys and failed. Commit
- 🔢 Empty embedding batch size. Clearing the Embedding Batch Size in Admin Settings > Documents now saves the default of 1, where files uploaded afterwards failed to process. #33288, #33287, Commit
- 🔊 Read Aloud with Kokoro.js. Read Aloud now speaks with the chosen Kokoro voice when Kokoro.js (Browser) is picked in Settings > Audio, where it used the browser's built-in voice while the server's speech engine was Web API. #33211, #33210, Commit
- 🔉 Kokoro.js without a GPU. Kokoro.js (Browser) now runs on the CPU in browsers that offer WebGPU but have no GPU available, where its voices never finished loading and Read Aloud kept spinning without speaking. #33233, #33232, Commit
- 🎨 ComfyUI failures reported. A ComfyUI workflow that fails, for example when the GPU runs out of memory, now shows an error and tells the model the image could not be made, where it was reported as a successful generation with no images. #33237, #33236, Commit
- 🖌️ Responses API images shown. Images that a model creates through the Responses API, such as Gemini image models through OpenRouter, now show in the chat, where they were saved but not displayed. #33320, Commit
- 🛝 Playground Controls applied. Settings chosen under Controls in the Playground, such as Temperature and Seed, now reach Ollama models, where they were dropped and every run gave a different reply. #33286, #33285, Commit
- 📋 Shift paste in channels. Holding Shift while pasting long text into a channel message or thread reply now keeps it as text when Paste Large Text as File is on, as it already did in chats. #33190, #33189, Commit
- 🧵 Thread replies while loading. A thread reply or model answer that arrives while the thread is still loading now shows up in full, and a reply no longer shows twice. #33346, Commit, Commit
- 👥 Analytics model chats by group. A model's Chats tab in Admin Analytics now follows the User group filter, where it listed every user's chats with that model. #33207, #33206, Commit
- 🙈 Hidden files in folder counts. The item count under a folder in a chat's terminal file browser now leaves out hidden files while Show Hidden Files is off. #33213, #33212, Commit
- 📁 Expanded folders kept. Deleting, renaming or adding a file in a chat's Files tab now keeps expanded folders listed, where every expanded folder came up empty. #33195, #33194, Commit
- 📌 Pins kept in shared folders. A member's chat now keeps its pin when the owner of a shared folder moves it to another folder inside that folder, where it dropped out of the member's Pinned list. #33342, #33341, Commit
- 🌙 Overnight calendar events. A calendar event that ends after midnight, such as 8:00 PM to 1:00 AM, now keeps its end time on the next day, where saving dropped it and stored the event as ending at its start time. #33290, #33289, Commit
- 🗓️ Midnight calendar events. A calendar event that ends at midnight now shows only on its own day in Month view, where it also showed on the next day. #33292, #33291, Commit
- 🪝 Old webhook removable. A notification webhook carried over from before v0.11 now stays removed from Settings > Notifications, where it came right back after Remove. #33295, #33294, Commit
- 🔔 New default notification target. A notification target added after removing the last one now becomes the default, where it never did. #33284, #33283, Commit
- 🗣️ Translated search and Skills labels. Labels in the workspace search, the Skills list, the skill editor and the terminal file browser now follow the interface language, where several always showed in English. #33308, #33304, #33305, Commit
- 📜 Keyboard tab scrolling. Scrolling on the Settings > Keyboard tab no longer moves the whole Settings window. #33187, #33186, Commit
Changed
- 🔀 Model switch keeps your picks. Switching models in a chat now applies the new model's default tools, skills, filters and features while keeping the ones you turned on or off yourself, where every switch cleared your picks, and chats with several models now get the default tools and skills of every selected model. Commit, Commit
- 📣 Model mentions switch chats. Sending a message to a model picked with @ now switches the chat to that model, where the mention stayed above the input and the chat kept its own model. Commit
- 🔁 Regenerate uses current models. Regenerating a response in a chat with several models now uses the model currently picked for that column, where it reused the model that first wrote the response. Commit
- 📎 Note attachments in pages. Files attached to a note now sit inside the page where they were added, and the attached files of existing notes move to the end of the note's first page. Commit
- 📂 Folder moves keep chat dates. Moving a chat into or out of a folder now keeps its last activity date, so it stays sorted by when it was really last active, where it jumped to the top of the sidebar under Today. #33327, #30362, Commit
- 📮 Automation failure webhooks. Admin event webhooks set to System events only no longer receive automation run failures, which now go to webhooks set to specific users or groups that include the automation's owner. #33331, Commit
- 📃 Knowledge base API responses trimmed. Adding, updating or removing knowledge base files and renaming or sharing a knowledge base through the API now return only the affected files or an empty list, so these calls no longer slow down as the knowledge base grows, and the full list stays available from the paginated files endpoint. #33335, #31397, Commit
@ Every self-hosted release, weekly
One Monday email with the week's stable releases across the directory.
>_ Models for Open-WebUI
Open-weight models you can plug into Open-WebUI, most popular first.
More in AI
orbitSelf-hosted AI gateway connecting private data to local or hosted LLMs via one OpenAI-compatible API with built-in auth, quotas, audit logs, and admin UI.
OtariA self-hosted, open-source LLM gateway that unifies 40+ model providers behind one OpenAI-compatible endpoint, with virtual keys, budget controls, and usage analytics.
PaddleOCR LocalSelf-hosted document parsing workbench that converts PDF, images, and Office documents to Markdown using five OCR models with a WebUI, CLI, and Apple Silicon support.
pandaprobePandaProbe is a self-hosted, open-source platform for tracing, evaluating, and debugging AI agents with integrations for LangGraph, CrewAI, and Claude Agent SDK.
panySelf-hosted semantic search engine running in PostgreSQL via pgvector, using CLIP embeddings for text and image search to replace Pinecone or Weaviate.
paperclipSelf-hosted, open-source platform to manage AI agent teams with org charts, budgets, governance, and goal alignment.
PaperCortexSelf-hosted AI layer for Paperless-ngx adding semantic search, receipt extraction, and DATEV export via local Ollama.
Paperless-AI nextSelf-hosted AI document tagging for Paperless-ngx with OCR rescue, duplicate detection, history rollback, and local/hosted AI support.
paperless-iqPaperless IQ is a self-hosted AI add-on for Paperless-NGX adding metadata tagging, RAG search, vision analysis, and approval workflow — supports Ollama, Anthropic, OpenAI
PasteGuardSelf-hosted privacy proxy that masks personal data and secrets before requests reach AI providers like OpenAI and Anthropic.
PatchworkAn open-source CLI tool that automates pull request reviews, bug fixes, and security patching using your own LLM keys — runs locally or in CI/CD.
PetrichorPetrichor is an open-source, self-hosted platform turning Markdown into semantic wikis with agentic RAG for traceable, evidence-backed answers.
0 comments