Using the team in the CLI

Build and run your AI team from the terminal: the /team command, windows, subcommands and what you see during a run.

Your AI team works in the terminal too. The CLI reads and writes the same .peridot/team.json as the desktop app, so a team you build in one place is there in the other, and everything in How hand-offs work applies as it is: the same tools, limits, approvals and costs.

Building a team#

Type /team to open your team. It does not need an account to edit the team, but the agent only runs, and so only uses the team, when you are signed in.

╭─ Your AI team ──────────────────────────────────────────────────╮
│                                                                 │
│ Pick the lead or a bot to change it.                            │
│                                                                 │
│ ❯ Lead — hands work to Researcher, Tester                       │
│   Researcher — read · Explores the codebase and answers quest…  │
│   Tester — read · edit · Runs the tests                         │
│   + Add a bot                                                   │
│   Show the graph                                                │
│                                                                 │
│ ↑/↓ move · Enter select · Esc cancel                            │
╰─────────────────────────────────────────────────────────────────╯

Move with ↑ / ↓, choose with Enter, go back with Esc.

A bot's settings#

Setting What you do
Name Type a new name and press Enter.
Role One line the lead reads to decide when to use the bot.
Instructions Opens your editor ($VISUAL, then $EDITOR, else vi, or Notepad on Windows). Save and quit to keep your changes.
Model Pick one of the models your plan includes, or Same as the chat.
Edit files / Run commands Enter switches it on or off. Reading is always allowed.
Hands work to Check boxes: the bots this one may give tasks to.
The lead can use this bot Enter switches it.
Delete this bot Asks to confirm. Links to the bot are removed.

Every change is saved right away. There is no Save button in the terminal.

Subcommands#

Everything the windows do is also a command, handy for scripts and quick changes:

Command What it does
/team Open the team windows.
/team list The bots, what each can do, and whether the lead can use it.
/team graph The tree of hand-offs.
/team add [template|name] Add a bot from a template (researcher, reviewer, tester, writer), or a blank one with that name.
/team show <bot> All settings of a bot, including its instructions.
/team set <bot> <name|role|model|instructions> <text> Change one setting. For model, give a model id, or - for the chat's model.
/team access <bot> <edit|run> <on|off> Allow or stop editing files or running commands.
/team link <from> <to> Let <from> (lead or a bot) hand work to <to>.
/team unlink <from> <to> Stop it.
/team remove <bot> Delete a bot.

A bot can be named by its id or by its name, in any case. Examples:

/team add researcher
/team add tester
/team access tester run on
/team set tester role Runs the test suite and reports failures
/team link tester researcher

Working with the team#

Nothing special is needed: ask for what you want. When the lead decides to hand work over, you see it in the conversation:

◆ Lead → Researcher: Find where authentication is handled and list the files
◆ Lead → Tester: Run the test suite and report failures
● Researcher · Search(login)
  ⎿ Researcher · 14 lines
● Tester · Run(bun test)
  ⎿ Tester → Lead ✓ All 44 tests pass
  ⎿ Researcher → Lead ✓ Login is in auth.controller.ts and auth.service.ts
∷ Login lives in auth.controller.ts, and all 44 tests pass.

Approvals#

Edits and commands by a bot ask you first, like the lead's. The window title says whose request it is, for example Tester: Run this command?. When several bots ask at once, the windows come one after another. Choosing "Yes, and don't ask again" switches the whole CLI to auto approval, bots included. Ctrl + C stops the run, every bot that is working, and any approvals still waiting.

The graph in the terminal#

/team graph shows who may hand work to whom as a tree, with how many tasks each connection carried since you started the CLI:

Who may hand work to whom

Lead
├─▶ Researcher
└─▶ Tester (2 tasks)
    └─▶ Researcher ↻

▶ may hand work to · ↻ already shown above

Researcher        read
Tester            read · edit · run

The terminal has no moving graph. The conversation above is the activity log, and the desktop app's graph is there when you want the picture.

Differences from the desktop app#

Desktop CLI
Editing A page with a Save button. Windows and commands that save at once.
Seeing the team An animated graph and an Activity list. /team graph, and the hand-offs in the conversation.
Instructions A text box. Your editor.
Positions in the graph You drag dots. Not used. They are kept in the file as they are.
The team file, the limits, approvals, costs Same. Same.

Tips#

← Project instructions