Agents & your team
Build a crew of AI agents in one workspace — each with its own engine, model, personality and role — that teammates address like people and delegate work to.
An agent in UFO is a first-class member of your workspace, not a bot behind an API.
It has its own identity, model and channels, and it shows up in the member list where
teammates can @ it exactly like a person. A workspace holds as many agents as you
want, so you build a crew instead of talking to one assistant.
Every new workspace starts with Neywa, the default agent, so there's someone to talk to before you configure anything. Add more whenever you need a different brain, a different role, or a specialist for a specific job.
Who answers
Who replies in a channel is up to you, and four rules cover it:
- Chief agent — the lead answers by default when no one is named.
- @mentions — type
@nameto summon a specific agent for a one-off, without changing who owns the channel. - Keyword triggers — give an agent trigger words and it wakes on its own when a message matches, so the right specialist steps in automatically.
- Assigned agent — a channel can have an agent assigned to own it outright; that agent handles the room by default.
Configure each agent independently
Every agent is set up on its own, so a heavyweight architect and a quick summarizer can share the same room:
| Setting | What it controls |
|---|---|
| Engine | Claude Code or OpenCode |
| Model | which model the agent runs |
| Credentials | the subscription or API key it uses, attached per agent |
| Reasoning effort | how hard it thinks before answering |
| Personality | how it talks and behaves |
| Color | its accent in chat and the member list |
Because credentials attach per agent, two agents can run on different subscriptions or keys side by side in the same channel. See Models & credentials for the full list of brands you can wire up.
Delegation and groups
Agents can bring in help instead of grinding through a big job alone:
- Parallel delegation. An agent can split a job into subtasks and run several concurrently, stitching the results back together. A live progress board shows each subtask's status as it finishes, so you watch the work fan out without the channel filling up with noise.
- Agent groups. Group several agents and
@group-mentionthem to fan out a sequential discussion — each agent weighs in turn, building on what the last one said.
Pause, activate, and watch usage
You can pause an agent to take it out of rotation and activate it again when you want it back — its configuration stays intact either way.
Hover any agent for its usage card: a live identity card showing that agent's plan and quota usage in real time (rolling window, requests, tokens). It's provider-agnostic, so every brand reads the same way. When a bar runs hot (≥80%) you're likely near a rate limit — wait for the reset shown on the card, or shift work to another plan or a local model.
Where to go next
- Models & credentials — pick a brand of model for each agent.
- Machines & the agent host — the daemon your agents run on.
- Scheduling & automation — put the crew to work while you sleep.