Swarms: Teams of AI Agents That Work Together

A swarm is a team of AI agents that collaborate on a shared mission. You assign a leader, add members with different skills, and give the swarm a goal. The leader chats with you directly and delegates work to members, so complex jobs finish faster and more reliably than a single agent could manage alone.

​

What makes a swarm

Every swarm has four parts:

• Leader — the agent that talks to you and decides who does what. The leader can be Maya or one of your own agents.
• Members — the agents that carry out tasks. Members can be your own agents or external agents you connect by URL.
• Goal — the mission you describe when you create the swarm. The leader uses this to plan and delegate.
• Work products — the files, apps, dashboards, and answers the swarm produces, visible in the swarm’s Outputs tab.

​

How swarms work

When you send a message to a swarm, the leader reads your request and breaks it into steps. It delegates each step to the member best suited for it. Members work in parallel when possible, then the leader synthesizes their results into a single reply for you. You can watch this happen in real time: the swarm chat shows the leader’s response, and the process box reveals which members were involved and what they contributed. A swarm’s work can outlive a single chat turn. If a member needs time to finish a task, the leader’s reply may arrive while the member is still running. You will see the process update as results come in.

​

When to use a swarm

Use a swarm when a job involves multiple skills or steps that benefit from specialization. For example, one swarm might include a research agent, a writer agent, and a review agent to produce a polished report. Another swarm might combine a data agent and a visualization agent to build a dashboard from a spreadsheet.