cindy: an open-source AI agent client that orchestrates multiple models and harnesses to automate work across desktop and mobile

What it solves

Cindy provides a unified interface for AI agents to perform actual work on a user's local computer, browser, and phone. It eliminates the need to switch between different AI tools by integrating multiple models and "harnesses" (execution environments) into a single agent that can access real files and logged-in applications.

How it works

Cindy acts as a client-side orchestrator that can leverage various harnesses, such as Claude Code and Codex, to execute tasks. It allows users to mix and match models and harnesses mid-task, enabling a workflow where one agent might plan a task while others execute it in parallel and a third reviews it. The system supports local model integration, API keys, or a hosted cloud service. It also integrates with the Model Context Protocol (MCP) to connect to internal business systems and tools.

Who it’s for

It is designed for users who want a powerful, local AI agent capable of automating complex workflows across their desktop and mobile devices, as well as developers who wish to extend the agent's capabilities via plugins and custom skills.

Highlights

• Cross-Platform Execution: Works across desktop (Electron) and mobile (Expo/React Native).
• Multi-Harness Orchestration: Supports switching between different execution environments like Claude Code and Codex while maintaining continuous memory and tools.
• Extensible Architecture: Supports custom skills, memory correction, and MCP integration for internal tools.
• Flexible Model Access: Compatible with local models, personal API keys, or the official Cindy cloud service.