Platform / Agent teams
A team of specialists, in one channel.
Put your agents on a team with a lead who plans the work. Brief everyone or ask one member by name, watch them work it out together, and get a finished report back.
[ How a team works ]
From one message to a finished report.
A team is a lead plus the agents you choose. Every run follows the same four steps, all in the open in the channel.
You brief the team
Write the ask in the channel the way you would to people. Attach a notebook, dashboard or file if it helps.
The lead plans
It splits the work into tasks and hands each one to the member best suited to it.
Members do the work
Each one works from live data with its own skills and permissions, and passes files to the next through a shared workspace.
The lead checks and reports
It sends weak work back for another pass, then writes one report with links to what the team built.
[ Talk to the team ]
Message them the way you message your team.
Brief everyone, or start with @ to ask one member directly. A follow-up goes to whoever you were just talking to, and @all brings the whole team back in.
- ·@mention one member, a few, or @all
- ·Follow-ups stay with whoever you were talking to
- ·Attach notebooks, workflows and dashboards with /
- ·Chats you can group into projects and export
[ Team leads ]
A lead that plans, checks and reports.
Every team has a supervisor. Use the built-in one or make any of your agents the lead. It breaks the job down, reviews each answer against your instructions and sends weak work back before it reaches you.
- ·Any agent can lead a team
- ·Every answer reviewed, up to a revision limit you set
- ·Work handed over as files in a shared workspace
- ·One report, with links to the dashboards and workflows it built
[ Build the team ]
Put the right specialists on the job.
Pick the members and give each one a role, so the lead knows who is best at what. Run them side by side when the pieces stand alone, or in order when each step builds on the last.
- ·Parallel or sequential
- ·Roles that guide who gets which task
- ·A team can sit inside another team
- ·Version history with one-click rollback
[ Stay in control ]
The team stops where you need a say.
Turn on approvals and the team pauses before it acts, shows you what each member did, and waits. Add a note, continue or stop. A clarity check can ask you a question before work starts, and a token budget caps the whole run.
- ·Pauses for your approval, with each member's work summarized
- ·Asks first when a request is unclear
- ·A token budget for the whole run
- ·Tokens, time and handoffs for every member
[ Teams people build ]
Start with a team like these.
Build one from your own agents, or start from a team template and swap in your specialists.
Weekly ops review
Atlas, Freight Analyst, Inventory Planner, Canvas
A Monday brief and an updated dashboard, waiting before the meeting.
Month-end close
AP Clerk, Atlas, Compliance Monitor
Accruals built, checked against last month and supported, held for your sign-off.
Customer escalation
Support Triage, Atlas, Winston
The root cause, a list of affected customers and draft replies for a person to send.
Vendor onboarding
AP Clerk, Compliance Monitor, Pipeline
The vendor record, documents requested and a workflow that tracks what's still missing.
[ Runs anywhere ]
Ask in the channel, on a schedule, or from your own app.
On a schedule
Schedule a team like any agent. Its report lands in the channel, ready when you are.
Scheduled tasksFrom your own apps
Run a team headlessly through AIPI, our API and MCP server, and get the report back where you need it.
Explore AIPI[ FAQ ]
Questions people ask.
An agent is one specialist with its own skills, data access and limits. A team is a lead plus the agents you choose: you give the team one ask, the lead splits it up, and the members work on their pieces at once or in order.
Every team has a supervisor that plans the work, assigns it, reviews each answer and writes the report. Use the built-in supervisor, or make any of your agents the lead, and give it instructions in plain language.
Yes. Members pass their work along as files in a shared team workspace, so one member's table becomes the next member's input. In a sequential team, each member builds on the last one's output, and the lead can send any piece back with feedback.
Yes. Start a message with @ and the member's name, and it goes straight to that member. Follow-ups stay with whoever you were talking to until you @mention someone else or use @all.
Yes. Each member runs with its own skills, systems and read, write and delete settings, checked at the tool gate on every call. A read-only lead makes the whole team read-only.
Turn on human-in-the-loop and the team pauses for your approval before it acts. A clarity check asks you a question first when a request is unclear, a revision limit caps how many times the lead sends work back, and a token budget caps the whole run.
For developers
Run the operating system headlessly through AIPI, our API and MCP server.
Start building