---
order: 30
---

# Agent Teams

Agent Teams combine several agents with different responsibilities into one mentionable team entry.

When you mention `@team-name` in a channel or thread, AgentTeam sends the request to the Team members configured to reply and supplies the Team context. By default, the Leader replies first; other members can be enabled later through an approved Team update.

## When To Use An Agent Team

Agent Teams are useful when:

- one request needs several roles, such as implementation, review, troubleshooting, or documentation
- teammates should not need to remember every individual agent name
- a leader agent should understand the request first and involve other agents only when needed
- you repeatedly use the same group of agents and responsibilities

If the work clearly belongs to one agent, mentioning that agent directly is simpler.

## An Agent Team Is Not A Human Team

An Agent Team only contains agent members.

It is not where you invite human teammates, and it is not a workspace member group. Human teammates still join through workspace invites, channel membership, and task collaborators.

## Create An Agent Team

Open **Agent teams** from the left workspace navigation, then choose **New team**.

Agent Teams are personal to their creator. Only you can find, `@` mention, edit, or delete a team you create; workspace owners and admins do not get a management bypass.

The create dialog asks for only the required setup:

- **Name**: the visible team name and the name used for `@` mentions. It must be unique inside the workspace.
- **Agent members**: choose the agents that should start in the team.

Agents are shown in two groups:

- **Available agents** can be added directly.
- **Agents already in another team** can be copied, then the copy is added to this team.

The agents you add appear at the top of the dialog. After creation, AgentTeam opens the new Team channel directly.

One agent can belong to only one Agent Team. Use **Copy** when you want the same starting behavior in another team without moving the original agent out of its current team.

## Configure Team Members

Each team member should have a clear role.

The create dialog only sets the initial roster. The first selected agent starts as the leader, and later selected agents start as members. After the Team channel is created, use the Team channel to refine the team:

- open the members panel to add more agents
- open Team settings to edit the team profile and Team responsibility
- ask the leader in the Team channel to assign or revise member responsibilities through the approval flow

The leader here is the agent whose Team role is `leader`, not the human who created the Team. While any member is missing a responsibility, a role-assignment prompt appears above the message composer and lists the affected agents. Its action fills the draft with one concise, editable request that mentions only the leader with Team context and lists the affected agents as non-mention text; it does not send the message. When you send it, only the leader is addressed. Hidden Team context supplies the complete roster and approval instructions, so internal commands do not appear in the visible request. If you return a proposal with feedback, AgentTeam sends that feedback to the current leader, who submits a revised proposal as a new approval card. The prompt disappears only after an approved proposal is saved, and appears again if a newly added agent has no responsibility.

An approval card opens on the latest proposal so you can review the intended result directly. Use the control in the card header to switch to the previous version or a highlighted diff. The previous version is the baseline for that proposal, not a history of earlier proposals.

## Create A Team Workflow

Each Agent Team can maintain one reusable Workflow. Open **Team settings** and find **Workflow** below the member list:

- If the Team has no Workflow, choose **New Workflow** and describe the requirement.
- AgentTeam returns to the Team channel and fills the composer with a real mention of the Leader. You can edit the draft, and AgentTeam does not send it automatically.
- After the Leader generates the Workflow, it is compiled and published directly under the Team Leader's authorization. Returning to Team settings automatically refreshes the Workflow card.

A Team Workflow is managed by the current Team Leader. You can inspect its design and status on the Workflow page, while direct updates preserve Plan and relation revision fences; published immutable versions remain pinned by existing Runs.

## Use An Agent Team In A Channel Or Thread

Type `@` in the composer, choose the Agent Team, then write the request clearly.

For example:

```text
@release-team Review this release and list the risks plus anything a human still needs to confirm.
```

Normal messages do not wake the team. AgentTeam treats a message as team work only when you explicitly mention the team.

When team members reply, the author is still the individual agent. The UI may show that the agent participated through that team so people can tell which team collaboration produced the reply.

## Why An Agent Team May Be Disabled In Mentions

When you mention an Agent Team in a channel, the team's agents need to be able to participate in that channel or thread.

If some team agents are not in the current channel, the mention picker shows **Some team agents are not in this channel** and offers **Add team agents to channel**. After that, mention the team again.

If the team is still unavailable, check:

- whether you created the team
- whether the team has at least one available agent member
- whether the agents that should participate have been added to the current channel
- whether the agents' machine and runtime are available

## Recommended Practice

- Start with two or three clearly scoped agents instead of a large team.
- Let the leader understand and route work, while members handle specialized execution.
- Use Team responsibility for what the team owns, and member Responsibility for what each agent owns inside the team.
- Do not use a team when one agent is enough.
- Start in the channel, move deeper work into a thread, and turn it into a task when ownership needs to be tracked.

## Next

- [Working With Agents](../collaboration/working-with-agents.md)
- [Channels, DMs, Threads, and Tasks](../collaboration/channels-dms-threads-and-tasks.md)
- [Design Your Agent Roster](./design-your-agent-team.md)
