Open WebUI 0.13.1 adds coding agents to its terminal and introduces Tether, a desktop companion that connects a computer’s files, repositories and tools to Open WebUI. The release also brings cheaper tool chats, configurable tool permissions and a substantial round of workspace improvements.
Open WebUI Watch analysis
What happened
The release notes describe a broad update to Open WebUI. In terminals that offer agents, users can select Claude Code, Codex or Pi, chat or use voice, choose reasoning effort and service tier, and approve tool requests in the chat. Each conversation uses one agent; those chats cannot be regenerated, continued or edited and resent.
Tether is available for macOS, Windows and Linux. Once connected, a computer appears in the terminal menu and can be reached through Open WebUI in a browser, including on a phone. Open WebUI says the connection works outbound, without opening ports, and that terminal output resumes after a dropped connection.
The project’s v0.13.1 release notes also describe these practical changes:
Our top picks
- Bring coding agents into the terminal
Users can work with supported coding agents in chat or by voice, and approve their tool requests there. - Connect a computer with Tether
The companion app makes local files, repositories and tools available through Open WebUI, without opening network ports. - Cut the repeated prompt overhead in tool chats
Open WebUI says tool chats cost much less because repeated instructions and attached knowledge are no longer rewritten into every tool round, allowing provider prompt caching. - Add an optional Auto tool-permission mode
Administrators can enable a review model to allow tool calls it judges do not need confirmation. The mode is off by default; the review sends tool details, arguments and recent messages to its scoring API. - Include small files in full
Extracted text under the default 50,000-character threshold is included in full rather than retrieved in pieces. - Open large chats faster
The notes say large chats open about twice as fast, while some chat-list actions are four to six times faster. - Work across split panes and remembered tabs
Users can keep tabs side by side, with each chat’s tab layout and selected terminal saved for reopening.
Why it matters
This release brings AI agents closer to the files and tools people actually use, while making a choice about how much autonomy those agents get. Tether makes local access more convenient; Auto adds a review step between an agent and its tools. That review has a real data boundary, too: the tool call and recent messages go to the configured scoring API.
The update is not just an agent launch. Faster chats, file handling and split panes make Open WebUI a more capable workspace for people running models and agents through it. The release notes give concrete changes, though performance figures beyond those stated are not supplied.
Our read
The most consequential combination is Tether plus agent access: more useful work can happen in a person’s real environment, without pretending that permission design is a footnote. Auto is off by default, and that is a sensible starting point while administrators decide whether a review model belongs in their tool-permission chain. Anyone enabling it should understand what the scoring API receives.
What to watch
- Which coding agents and terminal setups prove most useful in everyday work.
- How administrators configure Auto and which scoring APIs they use.
- Whether Tether’s reconnecting sessions and local-file workflows hold up in regular use.
Discussion spark: Would you let a review model approve low-risk tool calls if it receives the tool arguments and recent chat messages, or should every action stay behind a person’s confirmation?
Sources and evidence
- v0.13.1 (8 November 2026, 11:07 UTC)
not affiliated with, endorsed by, or speaking for Open WebUI or its contributors