Every setting of a bot: role, instructions, model, access and hand-offs.
Open Your AI team from the sidebar. The page shows the graph on the left and a panel on the right. Everything about a bot is edited in the Bot tab of that panel.
Click Add bot at the top right and pick one:
| Template | Role | Access |
|---|---|---|
| Researcher | Explores the codebase and answers questions about how things work. | Read only |
| Reviewer | Reviews code or changes for bugs, security problems and style. | Read only |
| Tester | Runs the build and tests, and writes or fixes tests. | Read, edit, run commands |
| Writer | Writes and updates documentation, READMEs and comments. | Read, edit |
| Blank bot | Nothing filled in: you write the instructions. | Read only |
Templates only fill in the fields below. You can change everything afterwards. A new bot is saved immediately, selected, and added to the lead's list of bots it may hand work to.
On an empty team the graph also shows the templates as cards, so you can start with one click.
A team can have up to 12 bots.
Click a dot in the graph, or use the chips at the top of the Bot tab, to select a bot. The settings below appear. Changes are only written when you click Save (the button is disabled until something changed).
How the bot is shown in the graph, in the chat cards and in the activity list. Up to 40 characters. You can rename a bot at any time, because links between bots use an internal id that does not change.
One line that says what the bot is good at. This is the most important field after the instructions: the lead reads it to decide when to hand work to this bot.
| Weak | Strong |
|---|---|
Helper |
Finds where something is implemented and reports file paths with line numbers. Read-only. |
Code stuff |
Reviews a set of changes for bugs and security problems. |
Up to 200 characters.
Which model the bot uses. Same as the chat (the default) uses whatever model you picked in the chat. Otherwise choose one of the models your plan includes. See Models and effort. The effort level always follows the chat.
The bot's own system prompt: how it should work and what its report should look like. Up to 4000 characters. Peridot already tells every bot that it is a teammate, who gave it the task, how its tools work and that its final message is its report, so your instructions can focus on the craft:
You are a strict code reviewer. Look for bugs, edge cases and security
problems. Report a short list of findings, most serious first, each with
file:line, what is wrong and how to fix it. If everything is fine, say so
in one sentence. Do not change any files.What the bot may do in your project:
| Setting | Tools | Notes |
|---|---|---|
| Read files | list_files, read_file, search |
Always on. A bot that cannot read cannot do anything useful. |
| Edit files | write_file, edit_file |
Each edit still asks for your approval, unless you use Auto approval. |
| Run commands | run_command |
Each command still asks for your approval, unless you use Auto approval. |
A bot can only use the tools you ticked here. Even if its model tries to call another tool, the call is refused. Give each bot the least access it needs.
Who can give work to whom:
Each allowed hand-off is a line with an arrow in the graph.
Click Lead in the graph or in the chips. The panel lists every bot with a checkbox: tick the ones the lead may hand work to. This is saved right away. If you untick everyone, the lead loses the delegate tool and works alone, while your bots stay saved.
Select it and click Delete bot, then confirm. Links to it from the lead and other bots are removed.
Everything is saved in .peridot/team.json inside the project folder, which you can read, edit by hand, commit and share. See The team file. Positions you drag dots to in the graph are saved there too.