AI in your workspace
There are two ways to put AI in an Open WebUI Computer chat:
- Connect a model by API: OpenAI, Anthropic, Ollama, OpenRouter, or any OpenAI-compatible endpoint. Add it in Settings → Admin → Connections. See Connect a model.
- Plug in a coding agent you already subscribe to: Codex, Claude Code, Cursor, Grok, OpenCode, Cline, or Pi run as native chat backends using the CLI login already on your machine. No API key needed. See Use your coding agent subscription.
Both show up in the same model selector and work in the same workspace chat. You can mix them: an Ollama model for quick questions, your Claude Code subscription for real work.
| API connection | Coding agent backend | |
|---|---|---|
| Credentials | API key (Ollama accepts any text) | The CLI's existing login on the host (your subscription) |
| Who executes tools | Computer's built-in tools (files, commands, browser, search) | The agent's own CLI, running on your machine |
| Session resume | Chats persist per workspace and reopen anywhere | Same, plus the backend's own session is resumed automatically each turn |
What the AI can do on your machine
Whichever route you pick, the AI works where your files actually live. It can:
- Read and edit files in the workspace, and search across the codebase
- Run shell commands and read the output, foreground or background
- Browse the web (navigate, click, fill forms, take screenshots) and run web searches
- Spin up sub-agents for parallel work
- Run on a schedule: recurring tasks like "run tests every morning" (see scheduled tasks)
You control how much of that happens without you: see Approvals, plan mode, and review.
AI is optional
Everything else in Computer (files, editor, terminal, git, remote access) works with no model configured. Add AI when you want it, not before.