What Peridot is, where you can use it and how the pieces fit together.
Peridot is an AI workspace. You can chat with Claude and Gemini models, let an AI agent work directly in your code, and build a team of AI bots that split a job between them. It runs in the browser, as a desktop app, in your terminal and on your phone, all with the same account.
| What it is | Best for | |
|---|---|---|
| Chat (web) | The chat at tryperidot.io, in any browser. | Questions, writing, quick code help. |
| Peridot Code (desktop) | The desktop app. Works inside a project folder on your computer. | Changing real code, running tests, and AI teams. |
| Peridot CLI | The same coding agent in your terminal. | Working over SSH, scripting, keeping your hands on the keyboard. |
| Mobile apps | Android and iOS clients. | Chatting on the go. |
All of them use one account, one set of plans and the same models. Chats you have while signed in are saved and show up everywhere.
Models. You pick a model for every chat. Fast models answer quickly, larger ones handle hard problems. An effort setting trades speed for deeper thinking. See Models and effort.
The agent. In Peridot Code and the CLI, the model does not only talk. It can list, read and search the files of your project, create and edit them, and run commands, always confined to the folder you chose and, by default, only after you approve each edit or command. See Tools and approvals.
Your AI team. In the desktop app you can create bots, each with its own role, instructions, model and access. The agent you chat with becomes the lead: it can hand self-contained tasks to the bots, who work in parallel with their own context and report back. A graph shows who is working and how work moves between them. See the team overview.
Plugins. Small add-ons that give the desktop app and the CLI new commands and new tools for the agent. See Plugins.
You ── message ──▶ Peridot app ── one model turn ──▶ Peridot server ──▶ Claude / Gemini
▲ │
└──── text + tool calls ◀────────────┘
tools (read, edit, run) run here, on your computer, inside your projectThe server only talks to the model: it never touches your files. The apps run the agent loop locally. They ask the server for one model turn, run the tool calls the model asks for on your machine, send the results back, and repeat until the model has finished. That is why nothing outside your project folder is ever read or changed.