mirror of
https://github.com/open-webui/open-webui.git
synced 2026-11-21 06:56:42 +03:00
doc: changelog
This commit is contained in:
+12
-20
@@ -9,6 +9,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
|
||||
|
||||
### Added
|
||||
|
||||
- **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".
|
||||
- **Bring your own computer with Tether.** Tether 0.1.14 or later now brings what runs on your own computer to any Open WebUI you connect it to, from whatever device you open Open WebUI on: models from Ollama, LM Studio or any OpenAI-compatible server, OpenAPI tool servers, MCP servers (Tether 0.1.16 or later) and Open Terminal servers that you add in Tether show up just for you in the model selector, the Tools list of the + menu and the terminal menu, and Open WebUI reaches them through Tether, so their addresses, commands and keys never leave your computer. 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. This replaces Direct Connections. [#34397](https://github.com/open-webui/open-webui/issues/34397)
|
||||
- **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](https://github.com/open-webui/open-webui/issues/34167)
|
||||
- **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.
|
||||
- **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.
|
||||
- **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](https://github.com/open-webui/open-webui/pull/34064), [#34165](https://github.com/open-webui/open-webui/pull/34165), [#34147](https://github.com/open-webui/open-webui/pull/34147), [#34031](https://github.com/open-webui/open-webui/issues/34031)
|
||||
- **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.
|
||||
@@ -19,12 +23,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
|
||||
- **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.
|
||||
- **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.
|
||||
- **Self-service terminals.** With a Terminals server connected, users who have Write access to the connection can now create terminals under Terminals in the terminal menu of the message input, which shows for them even before they have a terminal, choose whether everyone shares one instance or each user, chat or model gets its own, set a start directory, a setup script whose progress and failures show in System info, the image, CPU, memory, storage, network profile and environment variables, share a terminal with users and groups, and start, stop, restart or delete its instances, and those who manage a terminal can also open the system info and files of other people's running instances.
|
||||
- **Your own connections through Tether.** Models of OpenAI-compatible connections, OpenAPI and MCP tool servers and terminals that you add in the Tether app on your computer now show up for you in the model selector, the Tools list of the + menu and the terminal menu, with Open WebUI reaching them through Tether so their addresses, commands and keys stay on your computer. [#34397](https://github.com/open-webui/open-webui/issues/34397)
|
||||
- **Chats remember their terminal.** Each chat now saves its terminal and working folder and switches back to them when you reopen it, including a terminal you turned off, while chats from earlier versions open without a terminal unless they are coding agent chats.
|
||||
- **Default terminal for all models.** Administrators can now pick a default terminal under Admin Settings > Models > Model Defaults, which new chats start with for every model that has no terminal of its own.
|
||||
- **Shared chat terminals.** People who can reply to a chat shared with Allow replies can now use that chat's own terminal in their messages, as the share dialog now points out. [#34103](https://github.com/open-webui/open-webui/issues/34103)
|
||||
- **Git panel issues and actions.** When the terminal supports it, the Git panel now also lists the repository's issues to browse, create, edit, comment on, close and reopen, the pull request view adds a status filter, activity, conflict status and a choice of merge method, searching issues and pull requests now searches the whole repository instead of only the loaded list, changed files can be picked with Ctrl or Shift click to discard several at once or all of them, and buttons such as Work on this, Fix, Address feedback and Resolve conflicts in chat write a ready-made request for the model into the message input.
|
||||
- **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.
|
||||
- **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](https://github.com/open-webui/open-webui/issues/34103)
|
||||
- **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](https://github.com/open-webui/open-webui/discussions/10344)
|
||||
- **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](https://github.com/open-webui/open-webui/discussions/13780), [#13927](https://github.com/open-webui/open-webui/discussions/13927)
|
||||
- **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.
|
||||
@@ -36,22 +36,15 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
|
||||
- **Faster profile pictures.** Large profile pictures of users, models and channel webhooks, as well as 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. [#34598](https://github.com/open-webui/open-webui/pull/34598), [#34597](https://github.com/open-webui/open-webui/issues/34597)
|
||||
- **Responsive knowledge exports.** Exporting a knowledge base no longer holds up other users' requests on the same server process while the archive is packed. [#34600](https://github.com/open-webui/open-webui/pull/34600), [#34599](https://github.com/open-webui/open-webui/issues/34599)
|
||||
- **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](https://github.com/open-webui/open-webui/pull/34602), [#29576](https://github.com/open-webui/open-webui/issues/29576)
|
||||
- **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.
|
||||
- **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, with diffs of files that other models edit through the terminal tools needing Open Terminal 0.17.20 or later.
|
||||
- **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](https://github.com/open-webui/open-webui/issues/34062)
|
||||
- **Reasoning effort for coding agents.** The reasoning effort set in the advanced parameters now also applies to Claude Code, Pi and the coding agents connected over ACP such as Cursor and Gemini, which stop with an error when the chosen level is not offered for the model. [#34167](https://github.com/open-webui/open-webui/issues/34167)
|
||||
- **More coding agents.** Hermes, Antigravity, GitHub Copilot, Goose and Qwen Code now work as coding agents, as do other ACP agents added to Open Terminal with "OPEN_TERMINAL_ACP_AGENTS", and choosing Auto for tool permissions now lets Codex, Claude Code and Grok approve routine actions themselves and pass on only what they escalate, with Open Terminal 0.17.19 or later for the new agents and for Claude Code and Grok.
|
||||
- **Regenerate in coding agent chats.** Replies in coding agent chats can now be regenerated and earlier messages edited, and a forked or cloned coding agent chat picks up where the original left off, with Codex and Claude Code continuing their own conversation from that point on an up-to-date Open Terminal and other agents starting over from the chat so far, while file changes the agent already made stay.
|
||||
- **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.
|
||||
- **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.
|
||||
- **Chat menu items moved.** Overview, Artifacts and Comments are no longer in the chat's menu and now open from the + menu of the tab bar, where Overview opens as a tab beside the chat instead of a full-screen window.
|
||||
- **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.
|
||||
- **View terminal files.** Terminals in the terminal menu of the message input now have a More menu with System info and a new View files option, which opens that terminal's files in a dialog without selecting the terminal for the chat.
|
||||
- **Live Files tab.** 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.
|
||||
- **Type a folder path.** Clicking the path above a Files tab now lets you type a folder to go to, absolute or relative to the current one, with Tab completing folder names, and with a file open it shows the file's full path to select and copy.
|
||||
- **Markdown outline in Files.** Markdown files shown in a terminal's Files preview now have an Outline button that lists their headings and jumps to the one you pick. [#34494](https://github.com/open-webui/open-webui/pull/34494), [#34525](https://github.com/open-webui/open-webui/discussions/34525)
|
||||
- **Terminal picker for tabs.** When no terminal is selected for the chat, the menu for opening a new tab now 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.
|
||||
- **Terminal tab titles.** Terminal tabs are now named after the title that the program running in them sets, or after the running program when the terminal reports it, and show an icon for well-known programs such as Claude Code, Codex, Python or Node.js.
|
||||
- **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](https://github.com/open-webui/open-webui/pull/34494), [#34525](https://github.com/open-webui/open-webui/discussions/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](https://github.com/open-webui/open-webui/pull/34085)
|
||||
- **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](https://github.com/open-webui/open-webui/pull/34000), [#34529](https://github.com/open-webui/open-webui/discussions/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.
|
||||
@@ -62,7 +55,6 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
|
||||
- **Clone arena models.** Each arena model in Admin Settings > Evaluations now has a Clone button that opens a new arena model filled in from it. [#34138](https://github.com/open-webui/open-webui/pull/34138)
|
||||
- **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.
|
||||
- **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.
|
||||
- **Open folder button.** Folders in the working directory menu of the message input now have an Open folder button that opens the folder in a new Files tab.
|
||||
- **Locked Docker dependencies.** The standard and slim Docker images now install every Python package except PyTorch from lock files pinned with hashes, so rebuilding the same version installs the same packages, and custom images that edit "backend/requirements.txt" must regenerate the locks with "backend/update-locks.sh" for the edit to take effect. [#33981](https://github.com/open-webui/open-webui/pull/33981), [#33979](https://github.com/open-webui/open-webui/issues/33979)
|
||||
- **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](https://github.com/open-webui/open-webui/pull/34583), [#34580](https://github.com/open-webui/open-webui/issues/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.
|
||||
@@ -160,15 +152,15 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
|
||||
|
||||
### Changed
|
||||
|
||||
- **Direct connections removed.** Settings > Connections and Settings > Tools are gone together with the Direct Connections and Direct Integrations switches in Admin Settings, the Direct Tool Servers permission and the "ENABLE_DIRECT_CONNECTIONS", "ENABLE_DIRECT_INTEGRATIONS" and "USER_PERMISSIONS_FEATURES_DIRECT_TOOL_SERVERS" settings, so models, tool servers and terminals that users added in their own settings no longer appear and have to be added in the Tether app instead.
|
||||
- **Terminals connections need templates.** Connections to a Terminals server no longer carry a policy in Admin Settings > Integrations, so the Policy ID, image, resource, storage, idle timeout and Terminal Contexts settings and the Refresh Terminals button are gone, the server has to be a Terminals version that supports terminal templates, and models, folders, automations and chats that used the old connection need one of the new terminals selected.
|
||||
- **Direct connections move to Tether.** Settings > Connections and Settings > Tools are gone together with the Direct Connections and Direct Integrations switches in Admin Settings, the Direct Tool Servers permission and the "ENABLE_DIRECT_CONNECTIONS", "ENABLE_DIRECT_INTEGRATIONS" and "USER_PERMISSIONS_FEATURES_DIRECT_TOOL_SERVERS" settings, so models, tool servers and terminals that users added in their own settings no longer appear. Users add them in the Tether app (0.1.14 or later, 0.1.16 for MCP servers) instead, where they stay on their own computer and work in every Open WebUI they connect.
|
||||
- **Terminals connections need templates.** Connections to a Terminals server no longer carry a policy in Admin Settings > Integrations, so the Policy ID, image, resource, storage, idle timeout and Terminal Contexts settings and the Refresh Terminals button are gone, the server has to run Terminals 0.4.0 or later, and models, folders, automations and chats that used the old connection need one of the new terminals selected. Upgrading Terminals from 0.3.x takes its manual migration, as there is no automatic one, new Terminals installations start without a template, Open WebUI now connects with the Terminals server key, and "TERMINALS_OPEN_WEBUI_URL" and "TERMINALS_TOKEN_CACHE_TTL" should be removed from the Terminals settings. Open Terminal and Tether terminal connections are not affected.
|
||||
- **Image settings become connections.** On the first start of this version, the image generation and editing settings are turned into image connections that everyone who could use them before keeps using, and from then on the image environment variables such as "IMAGE_GENERATION_ENGINE", "IMAGES_OPENAI_API_BASE_URL" and "COMFYUI_BASE_URL" are no longer read unless "ENABLE_PERSISTENT_CONFIG" is turned off.
|
||||
- **Image API model checks.** Image generation and edit requests sent through the API now run on the model they name, which has to be one of the caller's image models or the request fails with an error, instead of image generation ignoring the named model.
|
||||
- **New coding agent model IDs.** Coding agent models now get IDs made only of letters, digits, dashes and underscores, which makes pinned Cursor and Pi models work, and pins, default models and automations that point at an agent model whose name had other characters, dots included as in most Codex models, need the model picked again. [#34171](https://github.com/open-webui/open-webui/pull/34171), [#34168](https://github.com/open-webui/open-webui/issues/34168)
|
||||
- **Coding agent sessions restart once.** The first new message in a coding agent chat from an earlier version starts a new agent session that gets the chat so far as text, instead of resuming the agent's own session.
|
||||
- **Attachments menu renamed.** The Add Context entry in the + menu of the message input, which holds the options to attach web pages, files, notes, knowledge and chats, is now called Attachments. [#34431](https://github.com/open-webui/open-webui/pull/34431)
|
||||
- **Capabilities now Features.** The Capabilities section of the model editor is now called Features. [#34433](https://github.com/open-webui/open-webui/pull/34433)
|
||||
- **Terminal uploads go to files.** New terminal connections now start with Chat Uploads set to Filesystem and computers connected through Tether always use it, so files attached in a chat are saved into the selected terminal's working folder instead of being processed as regular chat uploads, while existing connections keep their setting.
|
||||
- **Terminal uploads go to files.** New terminal connections now start with Chat Uploads set to Filesystem and computers connected through Tether always use it, so files attached in a chat are saved into the selected terminal's working folder instead of being processed as regular chat uploads, while existing connections keep their setting. Open Terminal 0.17.20 or later is recommended, as it fixes uploads becoming unreadable in multi-user setups.
|
||||
- **Live update events renamed.** Real-time updates for notes, comments, the inbox, access changes, encrypted channels and channel app errors now arrive as typed messages on the main "events" socket channel, so custom clients that listened for the old event names such as "events:note" or "note:presence" need to listen there instead.
|
||||
- **Chat write access field.** Chat responses from the API now carry the saved sharing mode in "share_mode" and report whether you can reply in a new "write_access" field, so custom clients that read "share_mode" for that need to switch.
|
||||
|
||||
|
||||
Reference in New Issue
Block a user