Team overview

Build a team of AI bots that work on your project together, and see how they talk in a graph.

Your AI team lets you build a small team of AI bots for a project, and watch them work together. Instead of one agent doing everything in one long conversation, the agent you chat with, the lead, can hand self-contained parts of a job to specialists. They work in parallel, each in a fresh context, and send back a short report.

Note

Your AI team is available in the desktop app only. The CLI and the web chat use a single agent.

The idea#

                    ┌────────────┐
   you ── chat ───▶ │    Lead    │  the agent you talk to
                    └─────┬──────┘
          delegate(bot, task) │ ▲ report
        ┌──────────┬────────┴─┴──┬────────────┐
        ▼          ▼             ▼            ▼
  ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐
  │Researcher│ │ Reviewer │ │  Tester  │ │  Writer  │   your bots
  └──────────┘ └──────────┘ └──────────┘ └──────────┘

You only ever chat with the lead. There is no separate conversation with a bot.
A bot is a specialist you define: a name, a role, instructions, a model and what it may touch.
The lead decides, per message, whether to use the team. It may do small jobs itself.
Work moves in two directions: the lead gives a task, and the bot returns a report.

Why use a team#

Parallel work. Exploring the code, running tests and reviewing a diff do not depend on each other. The lead can give all three out at once.
Clean context. A bot starts empty and only sees its task. It can read twenty files without filling the lead's conversation with them. The lead only receives the report.
Focused behaviour. A reviewer with strict instructions reviews better than a general agent that is also trying to write code.
Right model for the job. Give an exploring bot a fast, cheap model and a reviewer a strong one.
Least privilege. A researcher can be read-only. Only the bots you allow can edit files or run commands.

What you get#

Part	What it is	Docs
Bots	Specialists you create from templates or from scratch. Saved in your project.	Creating bots
Hand-offs	The lead's `delegate` tool, and who may delegate to whom.	How hand-offs work
The graph	An Obsidian-style map of the team that lights up while bots work.	The graph and activity
The team file	`.peridot/team.json` in your project.	The team file
Patterns	Example teams, tips and troubleshooting.	Patterns and tips

A first look#

Open a project, then click Your AI team under New.
Click Add bot and choose a template, for example Researcher.
Select the bot in the graph, read its settings and click Save.
Go back to the chat and ask for something that fits: "Find where we validate emails and list every place."
The lead hands the search to the Researcher. You see a card Hand to Researcher: …. Open the card to read the report when it finishes.

Open Your AI team again while a run is in progress to watch the graph.

Good to know#

Each project has its own team. It is saved inside the project folder, so you can commit it and share it with your colleagues.
Approvals still apply. A bot that wants to edit a file or run a command asks you, exactly like the lead, and the card says which bot is asking.
A team costs more usage. Every model call by a bot counts like any other request. Use it where parallel, focused work pays off.
Nothing is hidden. Every hand-off appears in the chat and in the Activity list.

← Tools and approvals Creating bots →