Open WebUI’s 0.13.3 release lets users run coding agents through scheduled automations, with a saved terminal and working folder. It also brings practical fixes for chats, file views and tool use, making this a substantial update for people building AI workflows in Open WebUI.
Watch Desk analysis
What happened
The release notes date version 0.13.3 to 8 November. The headline change is in the automation editor: users can choose a terminal, working folder and coding agent, then schedule runs against that setup. Open WebUI checks that the terminal and agent are still available before a run begins, and sends agent results to a chat.
Other changes include switching terminal or working folder during an existing chat, and keeping open Files tabs in step when the terminal changes. The update also fixes DeepSeek thinking-mode errors in chats using a llama.cpp provider when skills, knowledge or memories are enabled.
Our top picks
- Schedule coding-agent runs
Automations can use a selected agent, terminal and folder, with results sent to a chat. - Change terminal or folder mid-chat
Users can switch either setting after sending messages, including in temporary chats. - Keep Files tabs in the right place
Switching terminals now updates open tabs to show files from the newly selected terminal. - Keep tool-call labels across replies
Models can see labels from earlier tool calls when continuing a reply or answering another message. - Search knowledge files with punctuation
Searches can now find literal punctuation, such as the full stop in “Art. 40”. - Make clipboard controls clearer on local HTTP
Clipboard-dependent prompts explain when access is blocked, and Auto-Copy stays off rather than failing silently.
Why it matters
These changes make Open WebUI’s agent workflows more usable, not merely more configurable. Scheduled tasks can be tied to the working environment they need, while the terminal and file fixes reduce the risk of a chat showing the wrong workspace. Small reliability improvements matter when an agent is meant to do more than answer a question.
Our read
The most consequential addition is the ability to schedule coding agents with a chosen terminal and folder. That gives users a more concrete way to delegate repeatable work, while the checks before a run add a sensible guard against a missing setup. The rest of the release is a tidy collection of fixes that makes the interface less likely to lose track of what the user meant.
What to watch
- Whether scheduled agent runs prove dependable across different terminals and folders.
- Whether users get enough control over what an automation can do when it runs.
- Whether future releases broaden agent automations beyond chat results.
Discussion spark: Would you trust a coding agent to run on a schedule in a chosen folder, or should every run require your approval?
Sources and evidence
- What It Really Takes to Build an AI-Ready Workforce | HumanX 2026 Agenda | San Francisco | April 6-9, 2026 (Publication date not supplied)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.