Build a custom AI agent that handles real work on your desktop and mobile
Open a browser, edit files, and pull tasks from your calendar to process them with an AI running directly on your desktop and mobile. Cindy is an open-source client that bundles multiple AI models and tools to complete actual project work. It is ready to use immediately after installation, and you can extend its memory and features to suit your preferences.
How it works in your environment
Cindy is installed on your computer and interacts directly with actual files and logged-in apps. The initially supported harnesses (frameworks for running agents) are Claude Code and Codex, with more being added. You can freely switch models and harnesses during a single task, while the workspace and memory are preserved. It is also possible for multiple agents to plan, execute in parallel, and review a single task with different combinations. It operates the browser, computer, and smartphone, and retrieves and processes work from IMs or calendars.
Parts you can modify
Being open-source, you can not only view the code but also modify it directly.
- Memory A single correction results in permanent learning, which is shared across harnesses.
- Skills Define working methods to reuse anywhere; team sharing features are under development.
- Automation Automatically schedule and execute repetitive tasks, then report the results.
- MCP Connect internal tools and business systems to the agent's access scope.
Model and account selection
The client is free to use, and the source code is published under the Apache-2.0 license. You can bring models in whatever way you prefer. Logging in to the official Cindy service allows usage to be deducted transparently, and you can authenticate existing Claude Code or Codex plans to use them without duplicate billing. You can also connect your own API keys or use local models. Selecting 'Skip Sign-In' without logging in allows you to use only local agents, while server-based features are disabled.
Development environment and structure
This repository consists of a pnpm monorepo containing desktop and mobile apps, along with shared packages. The desktop is developed with Electron, and the mobile with Expo and React Native. Node.js 22.x, pnpm 10.x, and Git LFS are required. Linux users should refer to the installation guides for Ubuntu, Arch Linux, and Omarchy. Backend services are in a separate repository and are not included in this monorepo.
Pre-use considerations
Cindy accesses actual files and apps on your computer, so caution is needed in environments handling sensitive data. Official distribution builds include TapDB usage analytics, which only collects device and OS metadata linked to the account ID. It does not collect chat content, file content, or working directory data. You can disable analytics when building from source. Model weights, datasets, prompts, etc., may be subject to separate license conditions and are not automatically covered by the Apache-2.0 license at the repository root. Security issues should be reported privately according to SECURITY.en.md rather than as public issues.
