Models & credentials
Pick a model brand — Claude, ChatGPT, Grok, GLM, OpenAI, Ollama, OpenRouter and more — and UFO wires it up; every agent can run a different one.
UFO asks one question: whose model do you want? Pick a brand from the picker and it handles the wiring — sign-ins, keys, base URLs, the lot. You never touch a config file, and every agent in your workspace can run a different brain, side by side.
The one thing worth knowing up front: there's nothing to configure to start. If Claude Code is already signed in on the host, Neywa works out of the box. The picker is there for when you want more.
Pick a brand
| Brand | How it connects |
|---|---|
| Claude subscription | Sign in once on the host machine; agents use that session directly. No keys. |
| ChatGPT & Grok subscriptions | Type !connect in chat — a sign-in link and device code arrive in the channel, and it finishes in the background. No keys. |
| Claude-compatible coding plans | GLM, Kimi, DeepSeek and MiniMax run on the real Claude Code, so they keep every capability — subagents, hooks, skills. |
| OpenAI | Bring an API key for GPT models. |
| Ollama | Local, remote, or Ollama Cloud — UFO manages its own Ollama, no separate install. |
| OpenRouter | One key, most of the market. |
| OpenCode Zen | OpenCode's curated model gateway. |
| OpenAI-compatible | Any endpoint that speaks the protocol. |
See Chat commands for !connect and the other bang commands.
Two engines, one chat surface
Under the hood, two engines drive your agents — Claude Code and OpenCode — and both normalize into the same chat stream. You can switch a channel's engine at runtime, mid- conversation, without losing the thread: the switch keeps session continuity, so the history and tools come along. Different channels can run different engines at the same time. Changing the AI is a toggle, not a redeploy.
Keys live at the workspace level
API keys are encrypted before they touch the database and stored at the workspace level, not per machine. Add a key once and every machine you pair can use it — run the same model from your laptop, a Mac mini, or a VPS without re-entering anything. See Machines for pairing.
Credentials attach per agent
Model and credentials attach per agent, so two agents can run on different subscriptions or keys in the same channel — one on your Claude plan, another on an OpenAI key, answering side by side. Give each teammate the brain that fits its job.
One usage meter
Every provider accounts for usage differently — Claude's usage endpoint, ChatGPT's quota, Grok's rate-limit headers, a GLM dashboard. UFO collapses all of them into one meter, so the workspace shows the same thing whichever brand an agent runs, and a bar that's running hot looks the same everywhere. Hover any agent to see its live usage.
Where to go next
- Agents & your team — give each agent its own model and role.
- Connect your tools (MCP) — the same tools follow agents on any engine.
- Chat commands —
!connect, engine toggles, and more.