Introducing Open WebUI v0.13.0: your model gets a computer. With Open Terminal connected, it opens apps on a real desktop, tests its work in a real browser, and leaves its changes in Git for you to review. All self-hosted, and you can watch it work or take the mouse.

It tests its own work in a browser. In a tab beside your chat, it builds a page, opens it, and checks it. In Chrome mode, you share one session. On sites you don't control, use Docker, keep Ask for approval on, and stay signed out of accounts you care about.

Review what it changed. Changes land in a Git panel as the model works. Open any diff, then commit, push, and work with pull requests without leaving the chat.

Make every tool call wait for you. Turn on approvals, and desktop and browser actions stop until you click Allow. An admin turns on Tool Permissions first, and the default is Full access.

Let small models take the simple questions. Put a large model in charge, and it can hand research, coding, and review to sub-agents on other models. They're off by default, and they have no limits unless you set them.

Bring your team in. Comment on the exact words of a chat or note, and the teammate you mention lands on that passage. An admin turns on Comments first.

Already running Open WebUI? Update as soon as you can. v0.13.0 includes security fixes and a database migration, so back up first. A few changes need a decision. Read the post:

Self-host Open WebUI today, for free: pip install open-webui open-webui serve You'll need Python 3.11 or 3.12. Or use the one-line Docker command in the guide. View the source: Get started, and tell us what you build:

Git as the audit trail is the real unlock here. My rule for self-hosted agents: no commit, no merge, and every terminal action logged before I trust it on my own server.

🔥🫶🏼🙏🏼

Awesome update 🙌

can the limits on sub-agents include a spend cap, or just turns and tokens?

A Git diff won't show what the browser uploaded or submitted. Those actions need their own log and approval before they run, even when the code changes are easy to revert.

I NEED TRANSLATING TO POLISH LANGUAGE !

leaving the changes in git for review is the part that makes this usable. how does it handle apps behind a login, does it get its own creds or borrow yours

Hello, Dear. I'm an influencer looking for possible Collaboration, if any campaign is ongoing, kindly hit me back... Thank you

Will the desktop side work with a local model too?

a computer for the model and git receipts for me? yes please

Giving the model a real desktop raises the stakes on where Open WebUI listens, an instance that opens apps and pushes to Git shouldn't sit on a forwarded port with only its login in front, how do most of you reach yours from outside?

The Git trail is the useful part: the live screen shows what the agent tried, the diff shows what it actually changed. When it misfires, you review the diff and revert instead of replaying the whole session.

Letting the model take the mouse on a real desktop is neat. Can you scope what it's allowed to open, or is it all or nothing once Open Terminal is connected?

Letting the model test its own work on a real desktop helps.

Que gran utilidad, habrá que probarla en cuanto la descargue

watching it work is the whole point. skipping the review loop and just watching the desktop is where the real trust gets built