Open WebUI 0.13.0, released on 3 November 2026, changes the app's database, includes security fixes the project has not yet described, and adds a tabbed workspace beside each chat. If you run Open WebUI at home with Ollama, do two things: copy your data before you update, and install 0.13.2 or later, not 0.13.0. Version 0.13.2 fixes a fault that left chats stuck on a loading spinner when Open WebUI was opened from another device on your network.
Open WebUI is a browser workspace: the page you chat in. The model itself runs in a model runner such as Ollama. This update changes the workspace, not your models.
The four releases so far
| Version | Released | What matters for a home setup |
|---|---|---|
| 0.13.0 | 3 November 2026 | Database change, security fixes, tabbed workspace, moved chat controls |
| 0.13.1 | 8 November 2026 | Coding agents in the model picker, Tether companion app, cheaper chats that use tools |
| 0.13.2 | 8 November 2026 | Fixes chats that would not load over plain HTTP at a LAN or Docker address |
| 0.13.3 | 8 November 2026 | Agents in scheduled automations; change the terminal or folder during a chat |
Before you update
Copy your data first. The 0.13.0 release notes say the update changes the database layout and recommend backing up the database and its related data before upgrading. An update that fails halfway can leave a database the older version cannot open.
If you followed our Docker instructions, use the back up, restore and update procedure in the Open WebUI installation guide. It copies the data volume and checks the copy before the container is replaced. Open WebUI's own backup instructions cover other installations.
If more than one copy of Open WebUI shares a database, update them all together. The release notes say updating one at a time is not supported for this release and will cause failures. A single container on one computer is not affected by that warning.
Update even if you do not want the new features. The release notes say 0.13.0 contains security fixes and recommend updating promptly. They do not list the fixes; the project says it will publish the advisories once CVE numbers (the public identifiers for security faults) are assigned. Until then, nobody outside the project can say how serious they are.
Why 0.13.2 and not 0.13.0
Many home setups open Open WebUI from a second device using an address such as http://192.168.1.20:3000. The 0.13.2 release notes describe a fault in exactly that case: over plain HTTP at a LAN or Docker address, new and existing chats stayed on a spinner with the error crypto.randomUUID is not a function. Comments and some Markdown editing failed the same way. Version 0.13.2 fixes both.
If you already updated and your chats will not load from a phone or another computer, this is the likely cause. Update to the latest release before changing any connection settings. If the page loads but lists no models, that is a different problem: see Open WebUI cannot connect to Ollama.
What moved
Two things you may look for are in new places:
- Chat controls. 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 beside the message box. The button at the top now opens the tabbed side panel.
- Overview. The map of a chat's messages and branches now opens as a full-screen window from Overview in the chat's menu.
Instructions written for earlier versions, including screenshots, may show the old positions.
What the tabbed workspace adds
The Tabs button at the top of a chat opens a workspace for files, terminal sessions, previews, voice calls and comments. Three tabs do the most:
| Tab | What it does | What it needs |
|---|---|---|
| Desktop | Shows a computer desktop that you and the model can both operate | An Open Terminal connection with desktop support, and a model that can call its desktop tools |
| Browser | Opens websites and local development servers beside the chat; in Chrome mode the model can drive the browser | Browser support on the selected terminal, and Chrome for the model's browser tools |
| Git panel | Review changes, stage, commit, manage branches, push and pull | A terminal folder that is a Git repository |
Open Terminal is a separate component that you connect yourself. Without it, the Desktop and Browser tabs have nothing to show.
Other additions: side chats (type /side or /btw) for a quick question that does not join the main conversation, comment threads that an administrator can switch on, a notification inbox, and limits on how many files a knowledge base may hold. Apps supplied by MCP servers can now appear in the chat in a restricted frame, and each tool call such an app wants to make waits for you to confirm it. For background, see how MCP connects a local model to files and tools.
Two defaults to change on a single computer
Version 0.13.0 loosens two limits. Both suit a server more than one desktop computer:
- Tool calls. One response can now run up to 1,024 rounds of tool calls, up from 256. Setting
CHAT_RESPONSE_MAX_TOOL_CALL_ITERATIONSto256keeps the old limit. - Sub-agents. The limits on sub-agents now default to none. They used to default to 20 at once, 20 in the background, 30 tool loops each and 30,000 characters of output. A model can also now hand work to other models you choose in the model editor.
On one home computer, every sub-agent is another request to the same model runner and the same memory. A second model loaded for a sub-agent needs its own memory as well. Set the sub-agent limits in Admin Settings before trying multi-agent work, and read how much RAM and VRAM local models need before loading two models at once.
Check where the new features send your data
Updating does not change where your existing chats are processed: the selected model provider still decides that. Some of the new features add routes off your computer:
- Coding agents as models (0.13.1). On an Open Terminal that offers them, agents such as Claude Code, Codex and Pi appear in the model picker. An agent uses whichever model service it is set up with, which for Claude Code and Codex is usually their makers' cloud service. Check that before pointing an agent at private files.
- The Browser tab opens real websites. Anything the model types into a page goes to that site.
- Tether (0.13.1) is a companion app that links your own computer to your Open WebUI server so models and agents can work in your files, and so you can reach them from another browser, including a phone. The release notes say its sign-in lasts until 90 days pass without use, by default.
The local AI privacy guide explains how to check each part of a workflow separately.
One alert was removed
Personal notifications no longer offer the "Retrieval Content process failed" alert, so a notification webhook will not tell you when a file fails to process. After adding a document, open it in the knowledge base and confirm it finished processing before you rely on answers about it.
Related reading
- How to install Open WebUI with Ollama
- Open WebUI: what it does and where your data goes
- Our earlier coverage of Open WebUI 0.11
- Open WebUI vs AnythingLLM