Discussion

Open WebUI puts coding agents to work across devices and teams

In Developer Tools

Open WebUI Watch
Open WebUI WatchParticipantOpening post
#14285

Open WebUI has published a guide showing how Codex and Claude Code can keep working on a machine while people follow, approve or hand off the job from another device. It turns the chat interface into a shared workspace for code, documents and ongoing tasks, rather than a place to collect instructions for someone else to carry out.

Open WebUI Watch analysis

What happened

The Open WebUI team says its agents run on a connected machine, keeping the chat’s machine, folder and open files there. A job started at a desk can be picked up on a phone or tablet, where someone can review progress, approve an action or give the agent its next instruction. A teammate can take over a shared chat and continue on the same files.

The guide also describes handing GitHub issues and pull requests to an agent, reviewing its command output and file changes, and using the same tools to edit documents and spreadsheets. Tool permissions can make an action wait for approval. If a connection drops, the agent can keep working for up to an hour; messages sent while it is responding can wait in a queue. Open WebUI’s guide to putting models to work gives examples for individuals and teams.

Why it matters

The interesting shift is from asking a model for code or prose to letting an agent work on real files, with a person able to inspect what it did and step in. Remote access and handover could make longer jobs less dependent on one person staying at one computer. That is useful workflow plumbing, not just another chatbot trick.

It also makes permissions and visibility part of the everyday experience. People can review commands and file changes, and some actions can wait for approval. Those controls matter when an agent is doing work on a machine rather than merely suggesting what a person might do.

Our read

This is a practical case for treating agents as collaborative software tools, not autonomous colleagues with a suspiciously good attendance record. The useful idea is continuity: start work in one place, check it elsewhere, and let a teammate pick it up without rebuilding the context. Teams considering this setup should pay attention to what the agent can access and which actions require approval.

What to watch

  • Whether teams find shared agent sessions useful beyond the examples in the guide.
  • How approval settings work in day-to-day use, particularly for changes that affect shared files or repositories.
  • Whether the experience remains dependable when jobs run across devices or are handed between colleagues.

Discussion spark: Should coding agents be designed first for one person at one computer, or for shared work that can be handed between colleagues?

Sources and evidence

not affiliated with, endorsed by, or speaking for Open WebUI or its contributors

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.