Introduction to agent teams

What an agent team is

An agent team combines a coordinator with several specialists to solve requests that require distinct capabilities. The user receives one response: the coordinator analyzes the request, consults the appropriate specialists, and consolidates their contributions.

Use a team when a task benefits from clearly separated expertise. A single agent is usually a better fit for one stable objective.

How a team is organized

Coordinator: interprets the request, selects participants, and writes the final response.
Specialists: solve only the part covered by their capability.
Routing instructions: explain when each specialist should participate.
Execution profile: limits participation, time, and usage.

The current release uses one coordination level: specialists report to the coordinator and cannot delegate to other agents.

What specialists can access

Each specialist receives the task required for its contribution and keeps its own resources. Joining a team does not grant access to another agent's knowledge, tools, credentials, or permissions.

Adding an agent to a team does not expand its permissions. Review each agent's knowledge and authorized tools before activating the team.

Routing and execution

The team can automatically select only the required specialists or consult every specialist. Independent contributions can run in parallel to reduce the total response time.

Studio provides three starting profiles:

Fast: prioritizes speed and lower usage.
Balanced: fits most coordination scenarios.
Exhaustive: supports broader collaboration at a higher time and usage cost.

Testing and traceability

Test realistic requests in Studio before using the team with end users. Each run shows who participated, why they were selected, what they contributed, how long they took, and the consolidated response.

The history also provides statuses, credits, quality, and aggregate metrics so you can improve routing rules with observable evidence.

When to choose another option

Use a single agent when one configuration can complete the task.
Use a workflow for a deterministic sequence of steps, conditions, tools, or approvals.
Use an agent team when combining specialist reasoning is the core requirement.