Agents and teams

Agents are ACP’s reusable execution profiles. Teams coordinate multiple agents under one orchestration layer when work should be split or specialized.

Agents and teams

Define reusable execution behavior with single agents or orchestrated teams.

What agents are

An agent in ACP packages a model choice, reasoning effort, instructions, skills, and execution identity into a reusable operating unit. Instead of re-explaining behavior on every run, you define it once and reuse it across projects and threads.

Give agents stable prompts, tools, and execution preferences.

Use profile identity to make role-specific behavior explicit.

Select agents directly on thread and task execution surfaces.

What teams add

Teams let one orchestrator coordinate multiple agents. This is useful when work should be decomposed into specialized roles like development, research, QA, or operations.

Use orchestrators to delegate work to role-specific agents.

Keep team behavior reproducible instead of ad hoc.

Model multi-agent workflows as a first-class ACP concept.

How agents fit into execution

Agents and teams can be assigned to tickets, chosen on threads, attached to projects, and paired with specific computers. That lets you encode both behavior and execution environment in a repeatable way.

Pair agents with the right computers and skills.

Track execution analytics per agent and team.

Promote common roles into reusable system agents.