makecindy/cindy

Cindy is an open source AI agent that runs on your own computer and is meant to actually finish real tasks, not just chat. Instead of building its own model from scratch, it brings together several existing AI harnesses and models, currently Claude Code and Codex, with more planned, and lets them work on your files and logged-in apps. It can also drive your web browser, control your computer, take instructions from your phone, and pick up requests sent through messaging apps or on a schedule. This repository holds the open source client for Cindy, meaning the desktop app for computers and the mobile app for phones, along with the shared code packages they both depend on, organized as a single project with multiple linked pieces. The client itself is free and its source code is released under the Apache 2.0 license. To actually run tasks it needs a model, and you have several options: sign into the official Cindy service and pay for usage as you go, connect an existing Claude Code or Codex coding plan you already subscribe to, bring your own API keys from a model provider, or use models that run locally on your machine. Because Cindy is open source, several parts of how she works can be changed by anyone. You can correct her memory once and she will remember it going forward across every harness she uses. You can teach her a particular way of doing a task once and reuse that skill repeatedly. Recurring work can be scheduled to run and report back on its own. Internal tools and business systems can be connected through MCP, a way of wiring outside tools into the agent. A plugin system for reshaping the app's features and interface is also in the works. To build Cindy from source you need Node.js version 22, the pnpm package manager version 10, and Git LFS for handling large files. There is also a hosted mode that requires a Cindy account, and a local only mode called Skip Sign-In that runs local agents without one, though that mode loses access to server-backed features. The client connects to Cindy's official cloud servers by default, so contributors do not need to run their own server to develop or test the app.