Disclosure: As an Amazon Associate I earn from qualifying purchases. This site contains affiliate links.
Open WebUI 0.11.1 Puts a Human Gate in Front of Tool Calls
open-webui tagged v0.11.1 on Aug 25, 2026 at 21:17 UTC. The headline is human-in-the-loop tool approval.
Key takeaways
- open-webui tagged v0.11.1 on Aug 25, 2026 at 21:17 UTC. The headline is human-in-the-loop tool approval.
- Where an admin turns it on, a saved conversation can pause each tool call for allow or deny, by button or shortcut. Automations, channel replies, and temporary chats are not in that gate.
- They also added a built-in ask-the-user tool, llama.cpp and LM Studio model load/unload from the admin dialog, and a delta streaming rewrite they say can cut append work by up to 1000x. That last number is theirs.
Local LLMs on NVIDIA Spark / ASUS GX10
Open WebUI tagged v0.11.1 this afternoon. Published Aug 25 at 21:17 UTC. The instrument is a GitHub release.
The useful line for a local box is the human gate. Where an administrator turns it on, a saved conversation can stop letting tools run freely and ask first. The model waits. You allow or deny one call at a time, by button or by keyboard shortcut. The choice sticks for that conversation. Switch back to free-running and anything already waiting is released. Automations, channel replies, and temporary chats are not in that loop. That is not a universal agent sandbox. It is an opt-in pause on the chats that keep state.
They also shipped a built-in tool so a model can put up to three multiple-choice questions in front of you, with room to type your own answer, and the question survives a reload in a saved conversation. Admins can now download, load, and unload models on llama.cpp and LM Studio connections from the manage-models dialog, and remove them on llama.cpp, next to the Ollama path that was already there.
Streaming is a rewrite. A reply now goes out as small deltas instead of resending the whole message so far. They say that cuts the data and the server work spent appending by up to 1000x on a very long reply, and that a reply still in progress can be picked up after a refresh if Redis is behind it. That 1000x sentence is theirs. I am not going to launder it into a number from this box.
The same notes recommend upgrading production. They say the release includes access-control fixes, and that some security details are withheld for a bit so admins can move first. I am not enumerating what they did not print.
I run Grok when it earns it, and a GX10 when I want the weights in the room. Open WebUI is the pane in front of that split. This tag is a gate and a stream rewrite. If I flip the approval switch on my own instance, that note comes next. Not this changelog.
Frequently asked questions
Yes. The GitHub release is tagged v0.11.1 and published Aug 25, 2026 at 21:17 UTC. I am reading their notes, not a bench I ran.
No. An administrator has to turn it on. Then you can switch a saved conversation to ask first. Switching back to free-running releases anything already waiting. Automations, channel replies, and temporary chats stay out of that gate.
They say a reply now streams as deltas instead of resending the accumulated text, and they claim up to 1000x less data and server append work on a very long reply. Resuming a reply in progress needs Redis. I did not measure it.
Affiliate Disclosure: As an Amazon Associate I earn from qualifying purchases. This site contains affiliate links.