Open WebUI 0.12.3 release notes, 16 Oct 2026

Added

🔐 End-to-end encrypted chats. Direct messages and private group channels can now hold a temporary end-to-end encrypted chat: a member picks Encrypted Chat in the channel menu, the other members see "Encrypted chat in progress" with a Join link, messages of up to 16 KiB are encrypted in the browser with MLS and only passed through the server without being saved, participants can verify each other by comparing codes and see when others are typing, and the chat is cleared when you leave or reload and ends when channel membership or access changes. Administrators turn it on under Admin Settings > General > Encrypted channels or with "ENABLE_ENCRYPTED_CHANNELS" and grant it with the new Encrypted Channels permission, off by default or set with "USER_PERMISSIONS_FEATURES_ENCRYPTED_CHANNELS", and it needs websocket connections with websocket server logging off and HTTPS in production. Commit, Commit, Commit, Commit
📡 Much lighter streaming with Redis. With Redis, a streaming reply is now stored under its own key and new text is appended to it instead of rewriting the whole reply into a shared hash on every update, and replies started on older instances are still read during an upgrade; with room channels on, each instance now only subscribes to the rooms of people connected to it, so Redis no longer sends it live updates for chats it does not serve. Commit Commit #32478
🔍 Compare view options and undo. The Compare revision view has a Comparison options menu to ignore whitespace, ignore emphasis markers or only show changes with the unchanged lines folded away, an Undo button that steps back through your edits in both panes in order, and changed lines that are now lined up by their matching text. Commit, Commit, Commit, Commit
📝 Richer Markdown file previews. Markdown files opened in the terminal file browser now render with the same renderer as chat replies, including math, footnotes, highlighted code blocks, alerts and Mermaid diagrams, as well as HTML blocks mixed with Markdown, laid out as a compact document, and task list checkboxes in an editable file can be ticked to save the change to the file. Commit Commit Commit #32514
👀 Live collaborators in shared chats. The top bar of a chat now shows the pictures of the other people who have the same chat open at that moment, up to three and a count for the rest, with a Collaborators list that opens their profile cards. Commit, Commit, Commit
✏️ Edit and regenerate in shared chats with replies. In a chat from a shared folder that allows replies, people who may edit messages can now edit a prompt to start a new branch and regenerate any reply, not only replies to their own messages, and others viewing the chat stay on the branch they had selected. Commit, Commit
🗄️ Folder access in the folder editor. Sharing a folder now happens through an Access button in the Edit Folder dialog instead of a Share item in the folder menu, the sharing mode can be set before anyone is added, and subfolders note that access granted to parent folders also applies to them. Commit, Commit
🔎 Workspace search. The search window now searches your chats, chats in folders shared with you, chats shared with you, channel and direct messages and notes in one place, with filters for content type, a folder or channel, and whether something was created by you or shared with you, highlighted matches, a preview of each result with Ctrl/Cmd + Enter, Copy link, and channel results that open the thread scrolled to the matching message. Commit, Commit, Commit
💭 Comments in document previews. Word, Excel, PowerPoint and PDF files opened in the terminal file browser now show their comments and replies, marked where they belong in the document and opened by hovering or tapping a marker, and older .xls spreadsheets can now be previewed too. Commit
🧹 Clear Input Selections shortcut. Clearing the model override, selected tools and filters and the web search, image generation and code interpreter toggles from the chat input is now a shortcut you can change in Settings > Shortcuts, still Escape by default. Commit
🙋 Ask about a file from the file browser. Selecting text in a file open in the terminal file browser now shows an "Ask or edit…" box that sends your request to the model with the file attached and the selection quoted, so it can answer or edit that file in the same terminal; unsaved edits can be saved before sending, and requests sent while a reply is running wait in the queue. Commit
☁️ Google Drive access limited to picked files. The Google Drive picker now asks only for access to the files a user picks instead of read access to their whole Drive, so self-hosted installs no longer need Google's restricted-scope verification and users no longer see the "Google hasn't verified this app" warning. #32531 Commit #29537
📜 Full text of long status lines. Status lines above a reply, such as a token usage summary sent by a function, can now be clicked to show their full text, and with several status lines the history shows every entry in full. #32532 Commit #30056
📏 Skill read limits. Administrators can set how many characters of a skill the model reads at once and how long the preview of longer instructions is, under Admin Settings > General with "Skill Read Limit" and "Skill Preview Length" or the "SKILL_CONTENT_MAX_CHARS" and "SKILL_PREVIEW_MAX_CHARS" environment variables, which default to 20000 and 5000. Commit
🔄 General improvements. Various improvements were implemented across the application to enhance performance, stability, and security.
🌐 Translation updates. Translations for Portuguese (Brazil), Chinese (Traditional), Arabic, Bengali, Hebrew, Georgian, Punjabi, Slovak, Serbian, Urdu, Uzbek and Vietnamese were enhanced and expanded. #32440, #32482, #32501

Fixed

🧾 API key header requests in the audit log. Requests authenticated with an API key in the x-api-key header, as Anthropic SDKs send it, or in a header set with "CUSTOM_API_KEY_HEADER" now appear in the audit log under the key's owner, and failed attempts with a wrong key are logged without a user, where these requests were missing from the audit log. #32443, Commit
📚 Knowledge sharing kept on rename. A collaborator with write access who edits a shared knowledge base's name or description no longer turns off sharing they could not set themselves, such as Public. #32450, Commit, #32449
🤝 Prompt and tool sharing kept on save. A collaborator with write access who saves a shared prompt or tool no longer makes it private for everyone; sharing they could not set themselves is now kept. #32493, Commit, #32492
📄 PDF export of long notes and chats. Long notes and chats now download as PDF, where Firefox failed with "Canvas exceeds max size" and Chrome saved a black PDF. #32459, Commit, #18334, #26373, #19583
⏩ Continue Response with several models. Continue Response now continues the reply it was clicked under, where in a chat with several models it could extend another model's reply. #32456, Commit, #32455
🏷️ Tags of imported chats. Imported chats now show their tags and the tags appear in tag suggestions, where tags missing from the account were not created on import. #32454, Commit, #32453
🔖 Tags of archived chats. Searching the sidebar for a tag whose chats are all archived no longer deletes the tag, which lost its original name when the chats were unarchived. #32452, Commit, #32451
⌨️ Input methods in channels. Pressing Enter to confirm a word while typing with an input method, such as for Japanese, Chinese or Korean, no longer sends and clears a channel message before the word is in. Commit
🖱️ Scroll error after leaving a chat. Scrolling a chat right before opening Notes no longer logs a TypeError in the browser console. #32458, Commit, #32457
📐 Steady first turn in new chats. The latest message and reply in a new chat now keep their real height when the chat gets saved and the message list reloads, so the view no longer jumps on the first turn. Commit
💬 Sending in channels. Your own channel message now stays in place when the server confirms it instead of being redrawn, and the channel scrolls to your new message as you send it, where it could jump or keep the old scroll position. Commit
📂 Terminal folder in new chats. In a new chat, files you attach are uploaded to the folder open in the terminal file browser and the file browser stays in that folder once the chat is created, where uploads could go to the default folder and the browser jumped back to it. Commit
📁 Shared folder settings and access. Editing a shared folder's models, system prompt or files no longer resets its sharing mode, and a subfolder in a folder shared with write access stays writable when it is also shared directly with read access. Commit, Commit
🔑 OAuth token refresh with client_secret_basic. When "OAUTH_TOKEN_ENDPOINT_AUTH_METHOD" is client_secret_basic, token refresh now sends the client credentials in the Basic header like sign-in does, where providers that accept only Basic rejected the refresh and Open WebUI deleted the user's OAuth tokens, breaking model connections and tool servers that use them until the next sign-in. #32507, Commit, #32506
🛠️ OAuth tools after sign-in. An OAuth MCP tool now stays turned on after its sign-in redirect when the model has default tools, and links with ?tools= turn the listed tools on. #32488, Commit, #32476
🔤 Admin default font kept. Saving Settings > Interface no longer stores the administrator's default font and image compression size as your own settings, so later changes to those defaults still reach you. #32497, Commit, #32496
🔊 Read aloud with slow speech. Read aloud no longer stops after the first sentence when the speech engine takes longer to produce the next one than the current one takes to play. #32484, Commit, #32446
🖼️ Inline HTML previews on Per chat terminals. Inline HTML previews in replies now find the file in the chat's own terminal workspace when the Open Terminal chat context is Per chat, where they showed "File not found". #32485, Commit, #32477
🧵 Replies in channel threads. A reply started in one channel thread no longer carries over to the next thread you open, where messages could quote a message from the other thread. #32490, Commit, #32489
📑 Scanned PDF previews. Previewing a scanned PDF without a text layer in the knowledge base viewer no longer scrolls the page up and pushes the header and toolbar out of reach. #32487, Commit, #32481
↩️ Right-to-left lists. Bullets, numbers and task checkboxes of Markdown lists now sit on the correct side for right-to-left text in replies and file previews. Commit, #32444
🔁 Regenerate in cloned shared chats. Regenerate now works in a chat you cloned or forked from someone else's share, where it failed with "You can only regenerate your own messages". #32508, Commit, #32505
🗑️ Delete hidden for repliers in shared chats. People replying in a shared chat no longer see a Delete button the server refuses, which made a deleted reply come back on reload; Delete shows to the chat's owner and to administrators when "ENABLE_ADMIN_CHAT_ACCESS" is on. #32480, Commit, #32479
🎛️ Model default features can be turned off. A model can now switch off a Default Feature, such as Code Interpreter, that the administrator turned on in Model Defaults, where it came back on when reopened and in new chats. #32500, Commit, #32498
🧰 Built-in tools turned back on stay on. A built-in tool such as Memory that is off in Model Defaults now stays on after a model turns it back on and is saved. #32504, Commit, #32503
🏠 Refresh with Landing Page Mode set to Chat. Refreshing a new chat when Landing Page Mode is Chat no longer hides the message field and opens the model menu in the top left corner. #32510, Commit, #32509
🗜️ Compaction during tool runs keeps sources out. Context compaction during a long series of tool calls now summarizes your own message without the knowledge and file content added to it, and that content is given back to the model afterwards. Commit, #32491
🧩 Built-in tools with a custom tools parameter. A model with a custom parameter named tools, such as one that turns on a provider's native web search, now keeps its built-in tools, attached tools and file context, and its custom tools are sent alongside them. #32530 Commit #32528
🔌 Tool servers back after an outage. With Redis, tools from OpenAPI tool servers now come back once the servers are reachable again, where an empty list cached while every server was down kept them missing until a restart or a save in the tool server settings. #32533 Commit #29573
📥 PDF links read as documents. PDFs from sites that block unknown clients, such as Wikimedia, and arXiv links like arxiv.org/pdf/1706.03762 now go to document extraction, where the model got the raw PDF bytes. #32523 Commit #32524 Commit

Changed

🎯 Model skills now load their full instructions. With the Skills builtin tool on, skills attached to a model in the model editor, which every new chat pre-selects, and skills picked from the chat input menu now send their full instructions with each message, where they were only listed by name and description for the model to load with view_skill; deployments that attached skills to models only to make them available should remove them from the model to avoid the extra prompt length, since every skill the user can read is listed for the model anyway. Commit
📮 Redis permissions for room channels. With "WEBSOCKET_REDIS_ROOM_CHANNELS" on, the Redis user now needs the PUBLISH, SUBSCRIBE, UNSUBSCRIBE and PING permissions and access to the "&socketio" and "&socketio#*" channels, while PSUBSCRIBE is no longer used; a missing permission is logged with how to fix it. Commit