Teams
Give one mission to a group of agents. A coordinator decides how they work together, the members work in parallel where they can, and you get one deliverable back.
A team is a group of agents with a coordinator. You set the team up once in the Agency window, on the Teams tab. You run missions with it in the main chat, in Multi Agent Mode.
Create a team
- Open Agency from the sidebar, go to Teams, and press New Team. The team editor opens.
- Profile: enter a Name (required), pick a Category and describe the team's Purpose. Pick a theme gradient, or upload your own cover and avatar.
- Structure: press + next to an agent in the list to add it. Click a member, in the list or on the map, to Make lead, Remove them, and set who they Report to and Trust most.
- Coordination: keep Team runtime on, and write the Coordinator guidance: how the team should collaborate, challenge assumptions, use tools and decide the work is done.
- Limits: choose how many Members at once may work, and the Member time limit for a single step.
- Press Create team at the bottom of the rail.
A team needs a name and at least one member. When you edit an existing team, the members and their links save as you change them: the map reads Saves automatically, then Saved with the time. The name, category, purpose, theme and images save with Save changes.
The team editor
The team editor works like the agent editor: a rail of sections grouped under Team and Runtime, the team's cover and avatar on top with its member count and runtime on or runtime off, and the save button at the bottom of the rail.
| Section | What you set there |
|---|---|
| Profile | Name, Category and Purpose. A theme gradient for the cover and the avatar, or your own images, with Use gradient to go back. |
| Structure | The one place where members join and leave. Members shows who is on the team, with a star on the lead, and Find an agent to add searches everyone else. Each member shows its face, name and position. For the selected member: Make lead, Remove, Reports to, Trusts most, Memory search (Team, On or Off), Work mode (Team, Work or Off) and its Icon on the map, which replaces its face on the map. The Team map draws the members and who reports to whom; drag them to arrange it, and use the zoom and Full overview buttons under it. |
| Coordination | Team runtime: a coordinator assigns the work and members message each other until the mission is done. Coordinator guidance, which the coordinator turns into a protocol for each mission. Safety: Standard lets members message freely; Strict makes every message name its recipient and makes high-risk and critical actions wait for approval. Shared memory size: Fast (compacts the conversation at 300k tokens), Balanced (500k) or Deep (keeps up to 1M). Members start conversations lets members message each other without being asked. |
| Limits | Shown while Team runtime is on. Members at once: how many members may work at the same time. Member time limit: the longest one member may spend on a single step. Pick a preset or type an exact number. |
Removing a member also removes every Reports to and Trusts most link to or from them. If they led the team, the lead passes to the first remaining member.
The Teams tab
- Teams sit in categories: Engineering Teams, Growth Teams, Research Teams, Product Teams, Operations Teams, Customer Delivery and Uncategorized. Drag a card onto another category to move it. The pencil and the bin on a heading rename or delete a category.
- Each card shows the cover, avatar, name, purpose, category, member count, and Live or Offline. Its buttons are Team info, edit and delete.
- The toolbar has search (it also finds teams by member name), a filter (which also offers Live), a cards or table view, and a button for a new category.
- Team info shows the team's members and description.
You run missions from the main chat, described below.
Open a team in the chat
In the main composer, click Multi Agent Mode (the three linked circles), or open the picker's Teams tab and click a team. If you have no team, you see Create a team first.
Opening a team starts a first run, marked Ready. The composer now reads Message, then the team's name. Its first button shows the members' faces; click it to switch to another team.
The team page
The header shows the members' faces, the team's name, its status (Ready before the first mission), the member count and the run's name, which is the first mission you sent, or New run. Four views sit in the middle of the header: Run, Agents, Result and Setup. Your last choice is remembered in this browser.
Members and the coordinator
A strip of chips shows every member with a status dot. A crown marks the coordinator. Click a chip to make that agent the coordinator for new runs. While a run is paused, a click hands coordination over; while it runs, the chips are locked.
The mission status
At the top right, Mission tells you what the team is doing. Before the first message it invites you to describe the outcome and how the agents should work together; then it follows the run, for example Strategy ready or Team executing. Beside it you see how many members are working, or the run's state, and how long ago the run was created.
The six mission cards
Once a run exists, six cards show a live value and open a drawer when clicked:
| Card | What you set |
|---|---|
| Knowledge | Upload sources for this run: PDF, Word, Excel, PowerPoint, CSV, text, Markdown, JSON, HTML and OpenDocument files. |
| Questions | Clarifying questions that could change the mission. |
| Workforce | The persistent members, plus any temporary specialists you approved. Advanced: a simulated population the team can reason about. |
| Environment | Advanced and optional: a simulated environment with state you can edit and events you can inject. |
| Success | One observable criterion per line, up to 30. Press Save criteria. |
| Duration & Scale | Until success or Until manually stopped, a run time or end date, maximum rounds and cases, round timeout, how often to report progress, and how many checkpoints without progress to allow. Press Save limits. |
Send a mission
- Type the outcome you want in the composer and press Enter. Plain text only.
- The coordinator decides how to proceed. It may answer a simple request directly, ask a clarifying question, or propose a strategy.
- If it proposes a collaborative plan, a card marked Approval required shows the options, any temporary specialists, and the team graph. Press Approve, Revise (write what to change, then Send revision), or Cancel.
- Members receive their assignments and work, in parallel where possible. The coordinator adapts after each checkpoint. Long missions can sleep and wake on their own.
- The run finishes when your success criteria are supported. Open Result to read the deliverable.
Temporary specialists exist only after you approve them, and they are removed with that team session.
Watch the run
The Run view shows the conversation in time order. Your message appears at once with a chip (Sending, Accepted or Not sent). Then members' messages follow, with activity rows for thinking, tools and errors.
Your message shows Mission accepted while the team prepares its first contribution.
The Live team panel on the right lists each member with their face, current task or position, and state (Ready, Working, Done or Failed). Click a member to see their latest contribution. Hide the panel with its chevron and bring it back with the group icon. On narrow screens it starts hidden.
The toolbar has Resume (while paused), Pause and Stop (while running), Chat sessions for earlier runs, New run, Refresh, Add agents, and a button that opens Setup.
The Agents view
Under Agent chats, each member gets a pane with their face, name, position, model, the modes in use (RAG, Work, Deep) and a state. A pane shows that member's messages, tool steps, and the messages they exchanged with other members. You talk to the team through the main composer; the panes have no input box of their own.
- Focus pane enlarges one pane.
- Accent color changes that agent's color everywhere, not only in this team.
- Agent settings opens a Team member panel, saved to this team. In it you can swap the agent, pick a provider and model, set the reasoning level (Default to Ultra), toggle RAG, Work and Deep, or open Edit agent profile.
- Drag panes by their handle to reorder them. The order is kept in this browser.
The Result view
Result shows the final team deliverable, formatted, with Copy and Export (a PDF of the team conversation). Below it, the supporting contributions show which members added what. While work continues it reads Team is still working. With Open Run you go back to the conversation.
Team setup
Setup configures the team without leaving the chat, in four steps:
| Step | What you set |
|---|---|
| Outcome | The mission, success criteria and quality signals. |
| People | Add members; set each one's RAG and agent mode; make a lead; describe who they are, what they know, who they report to and who they trust most. |
| Collaboration | Protocols the coordinator can use: built-in Plan, Compare, Debate, Build and Review, or your own (a name, when to use it, and the protocol prompt). Also Coordinator guidance for new runs, and the switches Team communication, Shared inbox and Always preview. |
| Review | A summary, the team structure canvas, and advanced options. Press Save team. |
Always preview pauses even simple plans for your approval.
Commands and limits
- Type
/stop,/pause,/resumeor/statusin the composer to control the run. - Attachments are not accepted in team mode yet. Use plain text, and add files through the Knowledge card instead.
- The side pane closes while you chat with a team.
- Every member needs a model it can use. Set each member's model in the Agents view or in the agent's own editor.