# About Agent Pi Agent Pi is an AI coding agent that runs on your own computer. You talk to it in your browser or in its desktop app. It reads and edits your files, runs commands, uses a real browser, and can put a team of AI agents to work on one goal. It is in beta. ## Agent Pi in short | | | |---|---| | What it is | An AI coding agent you run yourself | | Where it runs | On your own computer: Windows, Mac or Linux | | How you use it | In Chrome at agentpi.app (with the Agent Pi extension), or in the desktop app | | What does the work | The oh-my-pi engine, installed for you. Little Coder is an optional second engine for AI models that run on your computer | | Which AI models | The ones you already have: Claude, ChatGPT and Codex, Gemini, DeepSeek, Grok, Mistral, OpenRouter and others, or local models through LM Studio and Ollama | | Status | Beta, by invitation | ## What you can do with it - **Give it a task in plain words.** "Add a login page", "find out why the tests fail", "tidy this folder". It reads the code, makes the changes, runs the commands and shows you each step. - **Put a team on a goal.** A lead, a developer and a reviewer, each on the model that suits the job, work in rounds until the goal is reached. Ready-made teams for development, SEO and video are included, and you can build your own. - **Let it use a browser.** With BrowserOS neo it opens sites, signs in with your own accounts, clicks through pages and reads them. - **Add tools when you need them.** Skills, extensions and MCP servers come from the [tools catalogue](https://ai-tools.agentpi.app/), where each one shows where it comes from and whether its code was scanned. - **Keep working across many chats.** Each chat has its own folder, model, tools and history. Long chats are compacted so they keep going. - **Run models on your own computer.** With LM Studio or Ollama, the model itself can run locally with no cloud at all. - **Automate.** Scheduled agents and n8n workflows run from the same place. ## How it works Agent Pi has three parts. 1. **The page.** The interface at agentpi.app, or the same interface inside the desktop app. 2. **The server on your computer.** A small program that runs the agent, keeps your chats and settings, and talks to the AI model you chose. It listens only on your own computer. 3. **The link between them.** In Chrome, the Agent Pi extension connects the page to the server on your computer. The desktop app does not need it. When you send a message, the page hands it to the server on your computer. The server runs the engine, which sends your request to the AI provider you picked and carries out the steps on your machine. ## Teams of agents A team is a set of agents with roles. One leads, others build, review or write. Each can use a different AI model and a different level of thinking. The lead reads what the others did, decides what happens next and says when the goal is met. If one member cannot answer, for example because its provider is down, the team treats it as absent and carries on. ## Which AI models it works with You bring your own account or API key. Agent Pi works with: - Claude (Anthropic), with a Claude subscription or an API key - ChatGPT and Codex (OpenAI) - Gemini (Google) - DeepSeek, Grok (xAI), Mistral, Qwen, Moonshot, Z.AI - OpenRouter, for many models behind one key - GitHub Copilot - Local models through LM Studio and Ollama ## What stays on your computer - Your chats, your code and files, and your API keys are kept on your computer. They are not sent to agentpi.app. - What you ask the agent goes to the AI provider you chose, directly from your computer, under that provider's own terms. With a local model, it goes nowhere. - The app fetches a few public files from the internet: its own updates, the tools catalogue, and the engines it installs. - The website keeps only what you give it: an email address if you ask for news, and a message if you use the contact form. Setup counts three anonymous steps (downloaded, started, finished). The full text is in the [privacy policy](/privacy). ## Who it is for - Developers who want an agent that works in their real project, on their own machine, with the models they already pay for. - Small teams and solo founders who want several agents to carry a job from plan to review without watching every step. - People who are not developers and want software work done by describing it. ## How to get it Agent Pi is in beta and shared by invitation. If you have a password, go to [Get Agent Pi](/downloads). If you do not, [ask for access](/contact). See [what is new](/changelog) and [what is planned](/changelog/planned). ## Questions ### What is Agent Pi? Agent Pi is an AI coding agent that runs on your own computer. You describe a task, and it reads your files, edits them, runs commands and reports back. It can also run a team of agents on one goal. ### Does my code leave my computer? Your code stays on your computer and is not sent to agentpi.app. The parts of it the agent needs for a task are sent to the AI provider you chose, as with any AI assistant. If you use a local model through LM Studio or Ollama, nothing is sent anywhere. ### Which AI models can I use? Claude, ChatGPT and Codex, Gemini, DeepSeek, Grok, Mistral, Qwen, Moonshot, Z.AI, GitHub Copilot, anything on OpenRouter, and local models through LM Studio and Ollama. You use your own subscription or API key. ### What do I need to run it? A computer with Windows, macOS or Linux. To use it in the browser you need Chrome and the Agent Pi extension. Setup installs everything else for you. ### Can it work without the internet? The agent can work offline when the AI model runs on your own computer through LM Studio or Ollama. Setting it up, updating it and using cloud models need the internet. ### What is the difference between Agent Pi and an AI assistant in a code editor? An assistant in an editor suggests code while you type. Agent Pi is given a goal and does the work: it plans, edits many files, runs commands and tests, uses a browser, and can split the job between several agents that check each other. ### What is oh-my-pi? oh-my-pi is the open-source coding agent engine that Agent Pi runs on. Agent Pi installs it for you and adds the interface, teams, tools and settings around it. Its changes are listed on the [changelog](/changelog/oh-my-pi) page. ### How much does it cost? The beta is free to use. You pay your AI provider for what you use with your own account or key. ### How do I get access? Agent Pi is in beta and shared by invitation. [Ask for access](/contact) and you will get a password for the [downloads page](/downloads).