Agents
Agents in Harmony work beside your team: they answer questions, read your connected systems, draft and change things, and take on tasks. Every agent works as a person, and nothing an agent writes to another system goes out without a yes.
The agents you can use
| Agent | What it is | Who it works as |
|---|---|---|
| Harmony Agent | Harmony's built-in agent. It is in every workspace and needs no setup. | The person who asked it |
| Your agents | Coding agents you run yourself, such as Claude Code or Codex, on your own computer or a Harmony computer. | You, with your own login and account |
| Team agents | Coding agents the workspace owns, running on the team's Harmony computers. Anyone in the workspace can use them. | The person who asked, using their own model key first |
Setting up your own agents and team agents is covered in External agents.
Where agents work
- In chat. Message Harmony Agent directly, or type @Agent in a room or workstream to ask it in a private side panel. See Asking Harmony Agent.
- On tasks. Assign a task to an agent the way you assign it to a person. See Agents on tasks.
- In the background. In a workstream, Harmony Agent can hand a long job to a background agent, which you can watch, steer and stop from the agents panel. See Background agents.
- On a schedule. A Loop runs the same instructions on a trigger. See Loops and runs.
Staying in control
- Before an agent creates, changes or sends something in a connected system, it asks you in the chat. See Approving what agents write.
- When someone asks your own agent to work on a task, you approve it first.
- Every run is recorded: what was asked, what the agent did, and what it cost. See Spend and models.