Skip to content

chore(deps): update ghcr.io/open-webui/open-webui docker tag to v0.13.4 - #297

Open
renovate[bot] wants to merge 1 commit into
masterfrom
renovate/ghcr.io-open-webui-open-webui-0.x
Open

renovate[bot] wants to merge 1 commit into
masterfrom
renovate/ghcr.io-open-webui-open-webui-0.x

Conversation

@renovate

@renovate renovate Bot commented Nov 18, 2026

Copy link
Copy Markdown
Contributor

Note

This PR body was truncated due to platform limits.

This PR contains the following updates:

Package Update Change
ghcr.io/open-webui/open-webui patch v0.13.3 → v0.13.4

Warning

Some dependencies could not be looked up. Check the Dependency Dashboard for more information.


Release Notes

open-webui/open-webui (ghcr.io/open-webui/open-webui)

v0.13.4

Compare Source

Added
  • Bring your own computer with Tether. Tether turns your own computer into part of Open WebUI: models from Ollama, LM Studio or any OpenAI-compatible server, OpenAPI and MCP tool servers and Open Terminal servers that you add in the Tether app now show up just for you in the model selector, the Tools list of the + menu and the terminal menu of any Open WebUI you connect, from whatever device you open it on, and Open WebUI reaches them through Tether, so their addresses, commands and keys never leave your computer. This needs Tether 0.1.14 or later, or 0.1.16 for MCP servers, tool servers and terminals need "ENABLE_TOOL_SERVERS" turned on, and a Tether model whose ID matches a model already in Open WebUI stays hidden. Tether replaces Direct Connections. #​34397
  • Sandboxed agents on your own computer. Tether 0.1.13 or later can lock the terminal on your computer into a sandbox, so models and coding agents work in your real files without reaching the rest of your machine: on macOS and supported Windows devices, Open Terminal and everything it launches, coding agents included, can only touch the one folder you pick, run offline unless you turn on Network access and accept no incoming connections, with no Docker, virtual machine or administrator setup needed. If the sandbox cannot be enforced, the terminal stays stopped instead of quietly running without it, and browser and desktop control stay off while it is on. Turn it on in Tether under Settings > Open Terminal > Sandbox.
  • Local MCP servers through Tether. Open WebUI can now use MCP servers that run as a local command over stdio, the way many MCP servers are distributed, by launching them on your computer through Tether 0.1.16 or later, with no proxy such as mcpo and no port to open. Add one in Tether under Settings > Integrations > Tools with its command, or connect an MCP server by its web address, and pick it from the Tools list of the + menu. Local commands run with your own account's access, outside Tether's workspace sandbox.
  • A computer for every user. With Terminals 0.4.0 or later connected, Open WebUI now hands out sandboxed computers on demand: anyone with Write access to the connection can create a terminal right from the terminal menu of the message input, pick whether everyone shares one machine or each user, chat or model gets its own, and set its image, CPU, memory, storage, network profile, environment variables, start directory and a setup script that, for example, clones a repository and installs its dependencies, without an administrator setting up each one. Terminals can be shared with users and groups and their instances started, stopped, restarted or deleted, a View files option opens any terminal's files without selecting it for the chat, System info shows how the setup script went, and those who manage a terminal can open the files and system info of other people's running instances. Start and Stop need Terminals 0.4.1 or later, the start directory and setup script need Open Terminal 0.17.7 or later in the terminal's image, and network profiles are the ones an administrator defined with "TERMINALS_NETWORK_PROFILES".
  • Coding agents as chat models. Coding agents now work much like any other model in the chat. Hermes, Antigravity, GitHub Copilot, Goose and Qwen Code join Codex, Claude Code, Cursor, Grok, OpenCode, Cline, Gemini CLI and Pi, and any other ACP agent added to Open Terminal with "OPEN_TERMINAL_ACP_AGENTS" works as well, once it is installed and signed in on the terminal. Auto for tool permissions now lets Codex, Claude Code and Grok approve routine actions themselves and pass on only what they escalate, and the reasoning effort in the advanced parameters now also applies to Claude Code, Pi and ACP agents that offer effort levels, such as OpenCode, which stop with an error when the chosen level is not offered for the model. Replies can now be regenerated and earlier messages edited, and a forked or cloned agent chat picks up where the original left off: Codex continues its own conversation from that point, other agents start over from the chat so far, and file changes the agent already made stay. Claude Code continues its own conversation too with the next Open Terminal release after 0.17.20 and until then asks you to update Open Terminal instead. The new agents and Auto for Claude Code and Grok need Open Terminal 0.17.19 or later, and effort for Claude Code, Pi and ACP agents needs 0.17.9 or later. #​34167
  • Pages open as tabs. Notes, channels, the workspace, automations, the calendar, the playground, the home page and new chats from pinned models now open as tabs beside the current chat when you click or drag them from the sidebar, pick them in the user menu or follow a link to one of them in a chat message, and starting a new chat, also from a sidebar folder, keeps the open tabs.
  • Queued messages steer replies. A message you send to your own chat while a reply is still running now waits on the server, so it is kept when you leave the page, and the reply wraps up at its next step so your message is answered right away with what it has done so far.
  • Instant first message. The first message of a new chat now shows up right away while the chat is still being created, instead of only once it exists, unless files are still being uploaded with it.
  • Full chat in side chats. Side chats now run the same chat as the main view, so message buttons, sources, token usage, follow-up suggestions, slash commands, queued messages, chat variables, web search settings, tool events such as code execution and input requests, HTML and SVG previews, notifications while the window is in the background, unsent text after a page reload and answers from several models at once now work there too, and a side chat's tab now has the chat menu. #​34064, #​34165, #​34147, #​34031
  • Interactive visualizations. Models can now draw interactive charts, diagrams, calculators and explainers right in the chat with the new Visualizations built-in tool, which is off by default and turned on per model under Builtin Tools, and administrators can rewrite its instructions under Admin Settings > Interface > Visualization Prompt or with "VISUALIZATION_PROMPT_TEMPLATE".
  • Video generation. Models can now generate videos with a built-in tool, also starting from an image in the chat, once an administrator turns on Video Generation under Admin Settings > Videos, adds connections to OpenAI, LiteLLM or OpenRouter, which offers models such as Veo and Kling, with a default video model, per-model defaults and access settings, and grants the Video Generation permission, which is off by default, after which users turn on Video in the + menu, can pick the model there and watch the result in the chat. #​13780, #​13927
  • Several image models. Administrators can now add several image connections under Admin Settings > Images, each an OpenAI-compatible server such as LiteLLM, Gemini, ComfyUI or AUTOMATIC1111 with its own models, sizes, workflows and access, used for generating images, editing them or both, where OpenAI-compatible ones offer the same sign-in options and custom headers as model connections and list the image models of a LiteLLM server by themselves when no model IDs are entered, choose the default generation and edit models and give a model its own default image model in its settings, and users with more than one image model pick one from the Image entry in the + menu of the message input. #​10344
  • Git panel becomes a GitHub workbench. With Open Terminal 0.17.8 or later and the GitHub CLI signed in on the terminal, as included in the full, agents and cuda images, the Git panel now covers the whole GitHub loop: browse, search, create, edit, comment on, close and reopen the repository's issues, see each pull request's checks, review state, labels, activity and conflicts with a status filter and a choice of merge method, and search issues and pull requests across the whole repository instead of only the loaded list. Buttons such as Work on this, Fix, Address feedback and Resolve conflicts in chat hand the issue, review or conflict to the model as a ready-made request in the message input, and changed files can be picked with Ctrl or Shift click to discard several at once or all of them.
  • Terminals follow your chats. Each chat now saves its terminal and working folder and switches back to them when you reopen it, including a terminal you turned off, and administrators can pick a default terminal under Admin Settings > Models > Model Defaults that new chats start with for every model without a terminal of its own, so every chat can start with a computer attached. Chats from earlier versions open without a terminal unless they are coding agent chats.
  • Folder chat board. Folder pages now list their chats with a search through titles and messages and filters for status, tags, people, activity and last update, and can show them as a board with columns and rows by status, tag or activity such as unread, generating or waiting for approval, gives each chat there the sidebar's chat menu to rename, move, archive or delete it, and a folder's badge in the sidebar now also counts chats waiting for approval. #​33771
  • Chat statuses in folders. Folders can now set up statuses for their chats under Edit statuses, with Backlog, To do, In progress, Waiting and Done offered as suggestions, chats then get a status from their menu on the folder page or by dragging them between board columns, people who can edit a shared folder can set the status of every chat in it, and each chat shows how long it has been in its status.
  • Folder automations. Folder pages now have an Automations section that lists the folder's automations with their state and last run and creates new ones there, chats made by an automation are marked with a clock, and automations can now post into shared folders their creator can edit, where everyone who can edit the folder can pause and run them and see their runs.
  • Knowledge on folder pages. Folder pages now have a Knowledge section that lists the knowledge bases, notes, chats and files new chats in the folder can use, where you can add, upload, drop, open and remove them.
  • Colored chat tags. Tags can now get a color on the folder page that applies wherever the tag is used, and the chat menu has a Tags submenu to find, add and create tags, and people who can edit a shared folder can tag every chat in it.
  • Edit automations in place. An automation's page now edits its name, instructions, status, schedule, model, terminal and destination right there with a Save button that appears once something changed, shows saved instructions collapsed behind Show more, lists its runs as activity below, and has Copy link and Duplicate in its More menu.
  • Readable agent and terminal steps. Steps of coding agents and the terminal tool calls of other models now show in the chat as what they did, a command with its output and its exit code when it failed, a diff of each edited file with its added and removed lines, or the files read, code and web searches, folders listed and pages fetched, and a group of steps is summarized as for example Ran 2 commands, Edited 1 file, Read 3 files. Diffs of files that other models edit through the terminal tools need Open Terminal 0.17.20 or later.
  • Models can mention people. With the Chat History built-in tool, models can now look up users by name or email and mention them in a reply, and each mentioned user who can see the chat gets an inbox notification that opens the chat at that reply.
  • OpenAI-compatible vector stores. External knowledge sources under Admin Settings > Integrations can now be OpenAI-compatible vector stores on OpenAI, Azure OpenAI or LiteLLM, searched on the server so no matching embedding model is needed, and you either pick one or more stores for a knowledge base or turn on 'List stores as knowledge bases' to show every store of the connection, or only chosen ones, as read-only knowledge bases that stay in sync with the endpoint and share the connection's access settings.
  • Review settings before saving. Admin settings pages now show next to Save how many settings changed since they were loaded or saved, and clicking that count or Save lists every change with its old and new value to confirm, with passwords and API keys kept hidden. #​34534, #​34532
  • Smoother replies with code blocks. Long replies with many code blocks no longer stall the tab a little longer with every new code block while they stream, so a reply with 300 code blocks that arrives at once now shows in 9 seconds instead of more than two minutes. #​34495, #​34492
  • Faster large PDF previews. Long PDFs in the file preview now open in about a second and use far less memory, because only the pages near the screen are drawn and pages far away are released, so a 1500-page manual no longer takes over half a minute and a large scanned file no longer pushes the browser past many gigabytes. The browser's find (Ctrl+F) now only matches text on pages near the screen. #​34602, #​29576
  • Faster folder reopening. Reopening a folder now shows its chats, details and coding agent models right away from what was already loaded and refreshes them in the background, sidebar folders keep their loaded chats when collapsed and expanded again, and folder chat lists update immediately when a chat is renamed, moved, archived or deleted.
  • Faster model list with Redis. With Redis, the model list now loads faster and Redis no longer keeps a copy of every model profile image, which helps most with large animated or imported images, and Functions and Tools now receive the model without its profile image. #​34595, #​34594
  • Faster profile pictures and knowledge exports. Large profile pictures of users, models and channel webhooks and knowledge base exports are now sent in one piece, so a 2 MB picture loads in about 15 ms instead of a few hundred and a page full of large avatars no longer takes seconds, and exporting a knowledge base no longer holds up other users' requests on the same server process while the archive is packed. #​34598, #​34597, #​34600, #​34599
  • Shared chat terminals. People who can reply to a chat shared with Allow replies can now work in that chat's terminal, as the share dialog now points out: with a Terminals template that gives each chat its own machine, and in coding agent chats, they work in the same machine and files as the owner, while with other terminals they use their own access to it. #​34103
  • Files tab upgrades. With a terminal that reports file changes, the Files tab now updates by itself without a loading spinner when files in the folders it shows change, also through commands run outside the chat. Clicking the path above it lets you type a folder to go to, absolute or relative to the current one, with Tab completing folder names, or select and copy an open file's full path, Markdown files get an Outline button that lists their headings and jumps to the one you pick, and folders in the working directory menu of the message input have an Open folder button that opens them in a new Files tab. #​34494, #​34525
  • Terminal tab upgrades. Terminal tabs are now named after the title that the program running in them sets, or after the running program itself with Open Terminal 0.17.8 or later on macOS and Linux, and show an icon for well-known programs such as Claude Code, Codex, Python or Node.js. When no terminal is selected for the chat, the menu for opening a new tab still offers Files, new files, terminals, the browser, the desktop and voice notes, and picking one first asks you to choose a terminal from a searchable list.
  • Grouped tools and skills. The Tools and Skills lists in the + menu of the message input are now grouped by where each entry comes from, such as the workspace, tool servers or the terminal. #​34085
  • Reply progress on tabs. The chat tab and side chat tabs now show a spinner while a reply is being written, so you can see from another tab when an answer is still coming.
  • Rename from the tab. Double-clicking the chat tab now renames the chat, also a chat someone else made in a folder you own, and on a folder page it renames the folder.
  • Share folders from their menu. The folder menu in the sidebar and on the folder page now has a Share option for administrators and for owners allowed to share folders.
  • Sub-agent permission. Administrators can now limit sub-agents to chosen users and groups with a Sub-agents switch under Features in the default and group permissions, or with "USER_PERMISSIONS_FEATURES_SUBAGENTS", and users without it get neither the sub-agent nor the timer tool. #​34040, #​34038
  • Lighter coding agent replies. While a coding agent writes its reply, the chat is no longer saved in full about once a second, so long agent chats write several times less to the database for each reply. #​34062
  • Labels on failed tool calls. Tool calls that failed now keep the short label the model gave them, also in a group of calls, where one failed call used to hide every label. #​34000, #​34529
  • Drop files into skills. Files and folders dragged onto a skill's files in the workspace now upload into the open folder, like the upload buttons do.
  • Editable calendars. Calendars can now be renamed and given another color after they are created, from a pencil next to each calendar in the Calendar sidebar. #​34136
  • Model names in analytics. Admin analytics now name coding agent models by agent and model for replies made from now on, and the tooltips of the session timeline show model names instead of model IDs.
  • Arena model management. The view filter in Admin Settings > Models now has an Arena Models option that lists only the arena models, and each arena model in Admin Settings > Evaluations has a Clone button that opens a new arena model filled in from it. #​34137, #​34138
  • OAuth session removals logged. Removing a stored OAuth session because of the "OAUTH_MAX_SESSIONS_PER_USER" limit, a sign-out sent by the identity provider or a reconnected MCP server now logs a warning naming the user, the session and the reason. #​34583, #​34580
  • Accessibility improvements. A tab moved with Alt+Shift+Arrow keys now keeps the keyboard focus, so it can be moved several places in a row.
  • General improvements. Various improvements were implemented across the application to enhance performance, stability, and security.
  • Translation updates. Translations for Azerbaijani, Basque, Bengali, Bosnian, Bulgarian, Cebuano, Croatian, Czech, Danish, Dutch, Estonian, Faroese, Filipino, Galician, Georgian, German, Greek, Hebrew, Hindi, Hungarian, Indonesian, Irish, Italian, Japanese, Kabyle, Korean, Latvian, Lithuanian, Malay, Norwegian Bokmål, Persian, Polish, Portuguese (Brazil), Punjabi, Romanian, Russian, Serbian, Slovak, Slovenian, Spanish (Spain), Swedish, Tamil, Thai, Tibetan, Turkish, Turkmen, Ukrainian, Urdu, Uyghur and Vietnamese were enhanced and expanded.
Fixed
  • Temporary chats leave no memories. The background memory review now skips temporary chats, so facts from them are no longer saved as memories that show up in later chats. #​34263, #​34262
  • Whole numbers in parameters. Whole-number fields in the advanced parameters, such as Context Length, Max Tokens and Seed, now reject a value with a decimal point, such as "32.767" typed for 32767, instead of saving it and running the model with a context of only a few tokens. #​34416
  • Busy servers with session sharing. With "DATABASE_ENABLE_SESSION_SHARING" on, a burst of page loads no longer stalls for 30 seconds and fails with a "QueuePool limit" error, because chat lists and the inbox no longer open a second database connection for their extra reads. #​34185, #​34184
  • Uploads during knowledge syncs. With a slow extraction engine, a large knowledge base sync no longer makes every other upload, chat attachments included, wait for an extraction before it is even saved, because files uploaded into knowledge bases now process at most half as many at a time as the server's worker threads ("THREAD_POOL_SIZE"), so other knowledge base uploads may wait their turn during a large sync. #​34379
  • Sign-in kept during provider outages. A sign-in provider that is unreachable, rate limited or failing with a server error while a user's token is refreshed no longer deletes that user's stored sign-in, so tools and connections that use the sign-in token work again on the next call instead of only after signing out and back in. #​34585, #​34578
  • Chat keeps the newest turn. A window that missed a turn, such as a laptop that just woke up, no longer moves the chat back to an older turn when it saves an edited message, so the newest reply stays in the conversation. #​34160, #​34126
  • Chats with looping history. A chat whose saved messages loop back on themselves now opens and shows its messages instead of freezing the browser tab until it crashes. #​34481, #​34477
  • Summary kept through compaction. A conversation compacted again between tool calls now keeps what its earlier summary held, and a chat compacted after such a compaction goes on from its newer summary instead of falling back to the older one and sending more than the model can take. #​34151, #​34141, #​34140
  • Steadier context compaction. Compacting a chat by hand right after a failed reply now keeps the summary for the next message instead of losing it along with the failed reply, and automatic compaction now also counts a tool result that ended a reply, such as one too long for the model to answer, so the next message compacts the chat instead of sending it in full. #​34434
  • Fewer repeated context lists. With native function calling, the list of available skills and the attached knowledge are no longer added to the chat again before every reply, only when they changed or after compaction, and recalled memories already in the chat are not sent again, so long chats use fewer tokens and need compaction later. #​34257
  • Tool call turns without text. Replies in which a model called a tool without writing any text are now sent back to the model without an empty text, so proxies such as LiteLLM no longer turn them into a placeholder line that the model then starts copying into its answers. #​34265, #​34264
  • Final answers on vLLM. When tool use stops, the request for a final answer no longer ends in a 400 error on vLLM, and Ollama models are kept from calling tools in it. #​34176, #​34169
  • Ollama proxy tool calls. OpenAI-compatible clients that send back an assistant turn with only tool calls and no text no longer get a 400 "Field required" error from the Ollama OpenAI-compatible chat endpoint, so multi-turn tool calling works through it. #​34442, #​34441
  • Timed-out tool calls explained. A terminal command or tool server call that runs past "AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER" now reaches the model as a timeout error instead of an empty one, so the model can tell you the call timed out. #​34119, #​34096
  • Tool calls in summaries. Conversation summaries written with the plain summary prompt, as for models with a filter, now include the tool calls of earlier replies in saved chats. #​34139
  • MCP tools reconnect. When an MCP server drops its connection during a reply, for example because it restarted, the next call to one of its tools now connects again instead of every later call in that reply failing with an "MCP client is not connected" error. #​34170
  • Large MCP tool lists. MCP servers whose tool list or a single tool result is larger than 1 MB connect and answer again, also through Tether, instead of failing with "Failed to connect to MCP server". #​34421, #​34201
  • MCP sign-ins with rotating tokens. Two tool calls that refresh the same MCP sign-in at the same time no longer delete the connection with providers that hand out a new refresh token on every refresh, because the refreshes now wait for each other, across workers when Redis is used. #​34582
  • MCP OAuth with HS256 tokens. Connecting an MCP server over OAuth 2.1 now works when the server signs its ID token with the client secret, as Readwise does, instead of failing with "Algorithm 'HS256' requires 'oct' key". #​34593, #​34504
  • MCP OAuth with large headers. Adding an MCP server with OAuth 2.1 and refreshing its tokens now work when its authorization server sends a very large response header, as monday.com does, instead of failing with "Connection error". #​34084, #​34078
  • Sign-in prompt for SSO tools. When the single sign-on session whose token a tool server forwards has ended, those tools now show as needing a sign-in in the + menu and picking one takes you to sign in again, instead of the tools failing on every call with no hint. #​34579
  • OAuth sessions on other devices. Signing in past "OAUTH_MAX_SESSIONS_PER_USER" now removes the stored provider sessions used least recently instead of the ones created first, so a device signed in early and still in use keeps its session, and signing out now removes that browser's stored session unless it is the person's only one for that provider. #​34586, #​34581
  • Quieter search logs. Knowledge searches and several web search engines no longer write found file names or full search results to the server log at the default level, only with "GLOBAL_LOG_LEVEL" set to "DEBUG". #​34146, #​34129
  • Deleted accounts leave access lists. Deleting an account now also removes it from the access lists of everything it was given access to, where it used to stay as a row showing only the deleted account's ID. #​34223, #​30023
  • Local speech-to-text on pip installs. Local speech-to-text in Open WebUI installed with pip now transcribes again instead of failing every recording with an error about "metadata_errors". #​34280, #​34278
  • OpenAI image generation defaults. Generating images with OpenAI now works with the stock settings, because GPT image models get the auto size instead of the old 512x512 default they reject and Auto can be saved with the model field left empty, and a failed request now shows the provider's own error, such as "Invalid size".
  • Fuller web page text. With the Playwright web loader, the text of definition lists, of figures with their captions and of collapsible sections, open or closed, now reaches the model, so attached web pages and web search results no longer lose content such as the function and parameter descriptions on many documentation sites. #​34282, #​34281, #​34277, #​34272
  • Unreachable sites skipped sooner. With SSL verification on, the Playwright, Tavily and Microsoft Web IQ web loaders now skip a site that never answers after 10 seconds instead of waiting about two minutes per link. #​34490, #​14241
  • Web pages behind internal certificates. With SSL verification on, the Playwright, Tavily and Microsoft Web IQ web loaders now load pages signed by an internal certificate authority set through "AIOHTTP_CLIENT_SSL_CERT_FILE" or "AIOHTTP_CLIENT_SESSION_SSL", and links that spell out a port or a username, instead of rejecting them. #​34482, #​34479, #​34462
  • No keyboard pop-up on phones. On a phone, opening a chat, switching chats or starting a new chat no longer puts the cursor in the message input and brings up the on-screen keyboard over the conversation. #​34059
  • Smooth Firefox scrolling. Scrolling up through a long chat in Firefox no longer throws the view back down, sometimes by several screens, while older messages load. #​33998, #​33994
  • Temporary chat choice kept. With Temporary Chat by default turned on in the interface settings, a new chat switched to a normal chat now stays normal after a page reload.
  • Permission changes reach open pages. When an admin changes permissions, open pages now follow without a reload, so a new chat that is open becomes a temporary chat once Enforce Temporary Chat is turned on instead of failing on the first message, Compare in the model selector turns off and only the first selected model stays once Multiple Models in Chat is turned off, an open channel shows the permission notice in place of the channel once Channels is turned off, and turning off Folders removes Move from the chat menu and closes an open Create Folder window. #​34513, #​34512, #​34515, #​34514, #​34511, #​34510, #​34509, #​34508, #​34507, #​34506
  • Images reach coding agents. Images attached to a message in a coding agent chat now reach the agent with the message instead of being dropped without a word. #​33993, #​33992, #​33991
  • Coding agent chat titles. Chats with a coding agent now get a generated title once the agent has replied, instead of keeping the default title. #​33971, #​33970
  • Agent model switches keep sessions. Switching a coding agent chat to another model of the same agent now resumes the agent's existing session and working folder instead of starting a new session.
  • Readable coding agent errors. When the terminal refuses to start a coding agent, the reply now shows a short error instead of a technical dump containing the whole system prompt. #​34173, #​34172
  • Sub-agents with browser questions. A sub-agent no longer hangs on "Executing Sub-agent" when a filter or tool asks the browser something, because the script, confirmation or form now appears in the tab that has the chat open. #​34045, #​34041
  • Sub-agent reports stay together. Two background sub-agent reports that finish close together no longer split the chat into two branches that hide the first report and the reply to it. #​34542, #​34540
  • Shared chat changes saved. Someone replying in a chat shared with Allow replies can now change its model, terminal and working folder, and the choice is saved to the chat instead of being dropped when it is reopened.
  • Terminal choice kept in folders. In a new chat inside a folder with a default terminal, a terminal you pick or turn off now stays that way when the model changes, instead of going back to the folder default, and a new chat in a folder without a default terminal keeps the terminal you had selected instead of turning it off. #​33973
  • Terminal locked during replies. The terminal and working folder pickers in the message input now stay locked while a reply is still being written, so the chat's terminal can no longer be switched in the middle of a reply. #​34383
  • Steadier Tether connections. Requests to a computer connected through Tether are no longer cut off after five minutes while data keeps flowing, read-only requests interrupted by a brief disconnect are repeated once Tether reconnects, and a busy computer now refuses new operations right away instead of letting some of them time out.
  • Your devices list updates live. A computer that connects through Tether after the page has loaded now appears under Your devices in the terminal menu right away, and the menu also picks up disconnects and device name changes, instead of needing a page reload.
  • Unsaved terminal file edits kept. An edit not yet saved in a file opened from a terminal's Files tab now comes back with its unsaved mark after you open another chat and return, and a workbook saved while you leave the chat keeps the worksheet you had open. #​34135, #​34107
  • Long Markdown files. Opening a long Markdown file from a terminal's Files, or typing in one of its code blocks, no longer freezes the page for many seconds, because code blocks without a known language now show as plain text instead of being checked against every language, which also keeps notes and the rich text message input responsive. #​34491, #​34486
  • File cards follow the terminal. A file shown inline in a reply now gets its preview, Download and Open back as soon as its terminal is picked again, and stops reaching the file once another terminal is picked, instead of keeping its first state until a page reload. #​34101
  • Working folder stays put. Browsing to another folder in a Files tab no longer changes the working folder shown in the working directory menu and used by a new chat, and opening Files again shows an existing Files tab where you left it instead of moving it to the working folder.
  • Terminal skill mentions after reload. A sent message that mentions a skill of the picked terminal now still shows it as a chip after a page reload, and the + menu lists that terminal's skills right away, instead of both only working after the skill list was opened once. #​34102
  • File sizes after Refresh. In a terminal's file viewer, Refresh and the back and forward arrows now keep the open file's real size instead of showing 0 B, and Refresh shows the new size when the file changed. #​34459, #​34458
  • Files menu near edges. The menu from a right click on the empty part of a Files list now opens upwards or to the left when there is no room below or beside the pointer, so all of its entries stay inside the window. #​34125, #​34097
  • Local embedding model folders. With the embedding model set to a folder on disk, saving Admin Settings > Documents no longer shows a "Model filesystem path detected" error and now also saves the embedding settings such as the batch size, which were skipped before. #​33978, #​33977
  • PowerPoint decks with sections. PowerPoint files that use sections now preview in the chat and the knowledge base file viewer and show their comments, instead of failing with "Missing slide relationship". #​34276, #​34270
  • Large workbook sheets. Excel workbooks with a sheet too large to preview now still open with their sheet tabs in a terminal's Files tab, the chat and the knowledge base, where that sheet says it is too large to preview and can be downloaded, instead of the whole file failing to preview. #​34247
  • Notes with footnotes. Notes containing footnote references such as "[^1]" now open and show their content instead of failing with an error. #​33987, #​33986
  • Note recordings on S3 storage. Recordings made in a note now upload to S3-compatible storage that rejects colons in file names and are named like "Recording-2026-11-16 15-25-06.webm" in every language, instead of after a date that slashes could cut short. #​34545, #​34543
  • Optional valves set to Custom. Switching an optional valve, such as an on/off setting, from Default to Custom in the Valves section of the chat controls no longer fails with "Error updating function user valves", and emptying an optional on/off or number valve sets it back to its default. #​34478, #​34475
  • Math in table headers. Formulas in table headers now keep their letters as written instead of being capitalized along with the header text. #​34245, #​34238
  • Math next to hyphens. Inline formulas written directly next to a hyphen or non-breaking hyphen, as some models write them, now render instead of staying plain text or showing a math error. #​34244, #​34237
  • Pinning keeps chat order. Pinning, unpinning, archiving or restoring a chat no longer moves it to the top of the chat list as if it had just been updated, and pinning a chat or moving it to a folder no longer marks it as read.
  • Drafts kept per folder. Unsent text in a new chat now stays with the folder or home page you typed it on, so opening another folder shows that folder's own unsent text instead of bringing yours along.
  • Links in messages. A link to a folder in a chat or channel message now opens the folder in Open WebUI instead of a new browser tab, and Ctrl or Cmd clicking a link to a chat, note or channel in a message now opens it in a new browser tab instead of the current one. #​34464
  • Tagging chats in parallel. Adding the same new tag to several chats at the same time, as an API client tagging a batch of chats or two chat imports can do, no longer fails with a server error for all but one chat and leaves the others untagged. #​34144, #​34143
  • Thread reply notifications. Clicking the browser notification for a reply in a channel thread now opens that thread instead of the channel with the thread closed. #​34233, #​34232
  • Side chats after reload. A saved side chat tab restored after a page reload or on another device now shows only the side conversation again, instead of also listing the messages copied from the parent chat. #​34002
  • Browser tab in two windows. With the same chat open in two windows, a Browser tab in Chrome mode no longer keeps reconnecting in both, and the window that lost the tab shows that the browser opened in another viewer. #​34133, #​34095
  • Ports open in new tabs. The Open in new tab button beside a port on the start page of a Browser tab now opens the server running on that port instead of a "404: Not Found" page. #​34120, #​34108
  • Smooth Browser tab resizing. Dragging the divider of the side panel no longer makes the Browser tab flicker and resize the page in slow motion. #​33989, #​33988
  • Terminal proxy headers. A "TERMINAL_PROXY_HEADERS" entry written in a different letter case now replaces the terminal's own header instead of being sent next to it, and repeated headers from the terminal such as several cookies are all passed on. #​34092
  • Groups need a name. A new group can no longer be saved without a name from a tab other than General, because Save sta

[!IMPORTANT]
✂ PR body was truncated to here.


Configuration

📅 Schedule: (in timezone America/Sao_Paulo)

  • Branch creation
    • At any time (no schedule defined)
  • Automerge
    • At any time (no schedule defined)

🚦 Automerge: Disabled by config. Please merge this manually once you are satisfied.

♻ Rebasing: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox.

🔕 Ignore: Close this PR and you won't be reminded about this update again.


  • If you want to rebase/retry this PR, check this box

This PR was generated by Mend Renovate. View the repository job log.

@renovate renovate Bot added the dependencies Dependency maintenance for the product, site or development tooling. label Nov 18, 2026
@github-actions github-actions Bot added the size: XS 0–9 changed lines label Nov 18, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

dependencies Dependency maintenance for the product, site or development tooling. size: XS 0–9 changed lines

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant