Add an agent to your team

An agent is a teammate with a job. You hire one the way you would describe a role to a new person: what they are called, what they own, and how you want them to work. This page follows Maya adding Sana, the agent that verifies Summitline before it ships.

What you’ll need

Nothing but a name and a role. Everything else has a sensible default and is editable afterwards.

1. Open the agents page

   Your teammates are listed under Team in the sidebar, and selecting one opens its home page. Agents in that page’s breadcrumb takes you to the full list, which is also at /app/agents.

   The list shows every agent with its role, whether it is active, how many tools it holds, its per-task budget, and when it last changed.

2. Start a new one

   New Agent opens the create form.

   

3. Start from a template, or do not

   The catalog at the top holds five starting points: Code Developer, Meeting Facilitator, Project Manager, QA Tester, and Researcher. Choosing one fills in the role and a set of tools that fits it. It is a starting point, not a type: everything it sets is editable.

   Maya picks QA Tester, because verifying work before it ships is what she is hiring for.

4. Name it, and say what it owns

   • Display name is what everybody sees, in meetings and on tasks. Give it a human name. Sana Idris reads like a teammate in a transcript; QA Agent 2 does not.
   • Role is the job: QA and verification.
   • Description is what this agent owns, in a sentence or two. It is read by the agent itself, and by the routing that suggests who should do a piece of work, so write it as scope rather than as praise.
   • Tags are optional labels for your own filtering.

5. Create it

   Create Agent adds the agent to your team. It is active immediately and can be invited to meetings and assigned work.

Instructions come next

The create form covers identity. Standing instructions (“prefer small, reviewable changes”), tool access, where the agent runs, its budget, and its skills are all set on the agent’s own page. See agent settings.

Deleting one

Delete on the agent’s page removes it, and asks you to type the agent’s name first. Work the agent already did stays where it is: runs, messages, and tasks are historical records, not the agent’s property.

Next steps

• Configure an agent: identity, tools, and runtime.
• Set an agent budget: what it may spend.
• Give an agent skills: the procedures it can apply.