Playbooks · Coding tasks
Open WebUI 0.13.3 puts agents in automations
Written November 16, 2026
Open WebUI 0.13.3 lets a scheduled automation pick a terminal, a folder and a coding agent, and checks both are still there before it runs.
Before this release, a self-hoster could chat with a coding agent from Open WebUI but could not put one on a schedule. Version 0.13.3, dated November 8, 2026 in the Open WebUI 0.13.3 release notes, closes that gap. The automation editor now works with coding agents, and the same release draws a firm line around where those agents can appear. Both halves matter: the scheduling is the new capability, and the boundaries decide what an agent automation can and cannot do once it exists.
What the automation editor does now
The release notes describe the editor in three steps. You choose a terminal, then a working folder, then a coding agent available there. The editor only offers an agent that the chosen terminal and folder actually provide, so the pick starts from what is present on that machine rather than from a general list of models.
A scheduled run uses the saved terminal and folder. Before it starts, it checks that the terminal and the agent are still available. That check is the practical part of the feature. A terminal can be disconnected, or an agent can go away, between the day an automation is set up and the day it fires. The run finds that out at the start instead of working from a stale setup.
Where the results go is narrower than for a plain model. Agent automations send their results to a chat. Channel destinations remain available for ordinary models, so sending a scheduled agent run into a channel is not part of this release.
Changing terminal or folder mid-chat
A related change applies outside automations. You can now choose a different terminal or working folder after sending messages, including in temporary chats. Both controls stay disabled while a reply is being generated, so the switch happens between replies rather than during one.
Two fixes keep that switching honest. Switching a chat to another terminal updates its open Files tabs to show that terminal's files, instead of leaving the previous terminal's folders on screen. With a terminal set to “Per chat”, the Files tab also switches to the chat's own files after the first message. And changing terminals immediately removes the previous terminal's agents from the chat's model list while the new list loads; a model-list request from a chat you have left can no longer replace the current list.
Where terminal agents must not appear
The release notes give this as a fix, under the heading of agents in unsupported model pickers. Connecting a terminal does not add its coding agents to any of the following:
- Admin Models
- base-model selectors
- Arena candidates
- task-model settings
- group defaults
- sub-agent settings
- channel mentions
- the Playground
Notes also stop choosing a terminal agent for title generation. Agents cannot be saved or imported as custom models, or as the base models for one, and an agent's chat menu no longer offers Edit. In other words, a terminal agent stays a thing you pick in a chat or an automation on that terminal. It does not become a model that the rest of Open WebUI can hand out, set as a default, or store as a custom model.
What to do with this
- Set the automation up on the right machine. Pick the terminal and folder first, then choose from the agents available there; the saved pair is what every later run will use.
- Expect results in a chat. If a scheduled job needs to land in a channel, that remains a job for an ordinary model in this release.
- Do not plan around defaults or mentions. A terminal agent cannot be a group default, a task model, or a channel mention, and it cannot be saved as a custom model.
- Keep the terminal connected. A scheduled run checks availability before starting, so a terminal that is gone at run time means the agent run does not start from the saved setup.
Related playbooks
- Zalando's broker hands agents OAuth safely
- Antigravity's remote agents run in Google's cloud
- Baseten Sandboxes keep agents running for days
- Cursor builds live apps from cloud agents
Sources
- Open WebUI v0.13.3 release notes — GitHub — agents in automations, changing terminal and folder during a chat, the Files tab and agent list fixes, and the list of pickers terminal agents stay out of.
Back to: Coding playbooks hub · Also: Copilot Automations learn from real outcomes