What open-webui/open-webui shipped
Written by FoxPlug from the repository's own releases; not affiliated with Open WebUI. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/open-webui/open-webui; Open WebUI did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.
Get a weekly update like this for your product, free
Week of October 5, 2026
What shipped
- Redis Sentinel now respects master name capitalization, fixing lookups that failed when the master name had uppercase letters like myMaster. Pull request #32234
- OAuth token forwarding and provider sign-out now stay valid after a browser restart, preserving the OAuth session cookies across restarts. Pull request #32209
- Automations, timers and sub-agents work again after the shared chats rework, fixing the Message already exists error that stopped replies. Pull request #32177
- MCP server connection failures now log at error level with the full cause instead of only at debug level, making troubleshooting easier. Pull request #32229
- Stopping a reply no longer leaves the chat stuck showing the Stop button and blocking the next message. Pull request #32184
- External knowledge citations from Qdrant, Milvus and pgvector now show the page title in the source list instead of the full address. Pull request #32232
- Creating a model in Manage Ollama now shows Ollama's error message when rejected and keeps the entered details instead of silently clearing the form. Pull request #32196
- Switching a Chat Controls parameter back to Default now uses the Settings value again instead of the model's built-in default. Pull request #32192
- Datalab Marker extraction now works on pip and source installs by saving extracted text to the upload folder instead of requiring the Docker image path. Pull request #32164
Why it matters
This week brought realtime voice calls to production and fixed critical issues blocking automations, OAuth sessions and chat replies. The fixes address reliability problems that broke core workflows and usability issues that confused administrators and users.
Changelog entry
- Fixed Redis Sentinel master name case sensitivity. Pull request #32234
- Fixed OAuth token forwarding across browser restarts. Pull request #32209
- Fixed automations, timers and sub-agents failing with Message already exists error. Pull request #32177
- MCP server connection failures now logged at error level with full cause. Pull request #32229
- Fixed stop button staying after reply was marked finished. Pull request #32184
- External knowledge citations show page title instead of full address. Pull request #32232
- Manage Ollama now shows rejection errors and preserves form data. Pull request #32196
- Chat Controls Default parameter now respects Settings values. Pull request #32192
- Datalab Marker extraction works on non-Docker installs. Pull request #32164
- Added Portuguese (Brazil) translations for newly added UI items. Pull request #32211
- Model editor language switching now refreshes fields to selected language. Pull request #32224
V0.12.0 ships with realtime voice calls and dozens of fixes: OAuth sessions survive browser restarts, automations work again, MCP errors show up in logs, and chat stops working as expected.
Open WebUI v0.12.0 is out with realtime voice calls for natural conversations, plus critical fixes for OAuth persistence across browser restarts, automations and timers getting replies again, MCP server connection errors showing up in logs, and chat reliability improvements. Dozens more fixes address admin settings, model management, citations and accessibility.
Week of September 28, 2026
What shipped
- Fixed chats becoming permanently broken when a reply was stopped or failed before any text was sent. Pull request #31892
- Stopped server memory from growing unbounded with every chat until restart by properly releasing locks. Pull request #31534
- Fixed large pastes into notes being lost after reload when text exceeded server size limits. Pull request #31893
- Fixed MCP OAuth sign-in failing with state errors when providers advertise many scopes. Pull request #31894
- Refactored Milvus hybrid search to eliminate duplicate text-only collections for better efficiency. Pull request #31660
- Fixed browser tab freezing when reopening chats with tool calls from providers that reuse response item IDs. Pull request #31838
- Fixed chats failing when MCP server ID plus tool name exceeds 64 character provider limits. Pull request #31822
- Preserved Chinese punctuation and quotes in uploaded file text extraction instead of converting to ASCII. Pull request #31655
- Implemented native hybrid search for Milvus that runs BM25 keyword scoring server-side. Pull request #31645
- Fixed Elasticsearch file edits leaving old text searchable by correcting chunk lookup logic. Pull request #31530
Why it matters
This week focused on stability and resource management, fixing memory leaks, chat-breaking bugs, and provider compatibility issues that could make the application unusable. Several improvements to knowledge base search and file handling ensure data integrity and correct retrieval, while MCP and OAuth fixes expand reliable tool integration.
Changelog entry
- Fixed chats becoming permanently unusable when a reply was stopped or failed before any text was sent Pull request #31892
- Fixed server memory growing unbounded until restart by releasing chat locks properly Pull request #31534
- Fixed large text pastes into notes being lost after reload when exceeding server size limits Pull request #31893
- Fixed MCP OAuth sign-in failing with invalid state error on providers with many scopes Pull request #31894
- Fixed browser tab freezing when reopening chats with repeated tool call IDs from some providers Pull request #31838
- Fixed chats failing when combined MCP server ID and tool name exceeds provider limits Pull request #31822
- Preserved Chinese punctuation and quotes in uploaded file text instead of converting to ASCII Pull request #31655
- Refactored Milvus hybrid search to eliminate duplicate text-only collections Pull request #31660
- Implemented native Milvus hybrid search with server-side BM25 keyword scoring Pull request #31645
- Fixed Elasticsearch file edits leaving old text searchable by correcting chunk lookup Pull request #31530
- Fixed iPhone HEIC photos being rejected by vision models in Firefox and browsers without type reporting Pull request #31649
- Fixed audit log masking passwords that contain double quotes or whitespace before colons Pull request #31659
This week: fixed server memory leaks, chats staying broken after failed replies, large paste losses, MCP OAuth issues, and improved Milvus hybrid search efficiency.
Open WebUI shipped fixes this week addressing stability and reliability. A critical server memory leak that grew unbounded with each chat is now resolved, chats no longer break permanently when replies fail early, and large file pastes no longer disappear after reload. MCP OAuth sign-in now handles providers with many scopes, Milvus hybrid search was refactored to eliminate duplicate collections, and file extraction preserves character integrity across languages.
Week of September 21, 2026
What shipped
- Admins can no longer bypass chat access restrictions through direct API requests when ENABLE_ADMIN_CHAT_ACCESS is disabled. Pull request #31416
- Model uploads and downloads now respect custom headers and authentication types for Ollama connections. Pull request #31490
- Manage Ollama dialog now applies custom headers and authentication types when listing, pulling, creating, copying and deleting models. Pull request #31489
- Tool approval mode now shows approval cards for all tool calls in a single turn, not just the first one. Pull request #31315
- File uploads and hybrid search now work with Qdrant strict mode by paginating reads in chunks of 1000 points. Pull request #31461
- Tavily web search now respects a configurable search depth setting instead of always using the default. Pull request #31308
- Searching automations and filtering models by non-ASCII words now works correctly on PostgreSQL with the default settings. Pull request #31423
- Chat header, folder, note and action menus now close before their dialogs open, so first clicks inside dialogs work. Pull request #31496
- The
max_tokensparameter sent through the API is now correctly passed to Ollama models as output length. Pull request #31437
Why it matters
This week focused on fixing access control and authentication issues that affected API security and multi-connection setups, plus resolving approval workflows and database compatibility problems. Several fixes address tool handling and parameter passing that were silently failing in production setups.
Changelog entry
- Security: Enforce chat access restrictions for admins through API requests Pull request #31416
- Fix: Model upload and download respect connection custom headers and auth Pull request #31490
- Fix: Manage Ollama applies custom headers and authentication types Pull request #31489
- Fix: Show approval cards for all tool calls in multi-call turns Pull request #31315
- Fix: File uploads and hybrid search work with Qdrant strict mode Pull request #31461
- Feature: Add configurable Tavily search depth setting Pull request #31308
- Fix: Non-ASCII search queries work on PostgreSQL with default settings Pull request #31423
- Fix:
max_tokensAPI parameter now reaches Ollama models Pull request #31437 - Fix: Dialog menus close before opening, first clicks work correctly Pull request #31496
- Fix: Spacing preserved in reasoning model answers after thinking block Pull request #31438
- Fix: Backslashes in uploaded HTML files handled correctly Pull request #31450
Weekly digest: access control, authentication, tool approvals, Qdrant compatibility, and API parameter handling all improved. 10 fixes for production stability.