Cindy

Cindy, from XD Inc., is positioned as an "out-of-the-box open-source AI agent." It connects your work computer, home computer, laptop and phone. When you switch devices, your tasks don't get interrupted. You can return to the same session or keep working on the same files.

Key features

• One Cindy across devices: Access the same tasks and conversations on multiple devices, and browse, preview and transfer task files. You can borrow subscriptions, APIs or local models already configured on another computer, so you don't have to set them up again on every device.
• Built-in remote desktop: Switch between connected computers from your phone, view the real screen, and take over control when needed.
• Multiple agent frameworks: Choose Claude Code, Codex or Pi in one place, while keeping your task history, memory and skills. Within the same task, you can hand off to a different agent.
• Multi-agent collaboration: Start with planning, then build, test and write documentation in parallel, and finish with an independent review. Each role can use a different framework and model.
• Bring your own models: Use the subscriptions or keys you already have, including logging in with Claude, ChatGPT, SuperGrok, Kimi Code, MiniMax or GitHub Copilot, or using your own API key. You can also connect local models such as Ollama, llama.cpp and LM Studio (Beta), or use Cindy's official hosted model service.
• Extensibility: Supports skills (SkillHub), a plugin marketplace and MCP, and can connect to tools such as Slack, GitHub and Feishu. SSH connections to server environments are supported.
• Control and traceability: Permissions are set per task, and requests that need confirmation wait for your approval, which you can also give from your phone. You can review code diffs and tool-call traces, and request an independent review. Eligible local Git tasks can use worktrees and checkpoints.
• Companion roles and routine tasks: You can define a name, responsibilities, memory and skills for each companion, and set up periodic checks and reports on your computer. You can also assign it new tasks through connected chat channels.

Target users and use cases

It suits developers who already use AI coding agents such as Claude Code or Codex and want to hand work off seamlessly between their phone and multiple computers. It's also a good fit for individuals who want to manage several model subscriptions and agents in a single interface. Teams can @Cindy in Slack or Feishu to have it join discussions and produce first drafts.

Core advantages

• Open source under the Apache-2.0 license, so you can audit, fork, build it yourself and contribute code.
• Tasks run on your own computer, and project files and full execution records are stored locally, so there's no need to rent or deploy a backend server.
• The software itself is free, and you can use your own keys or subscriptions without buying an additional model subscription.

Pricing and editions

• Free: Free of charge. Bring your own keys to use all of the software's features.
• Plus: From $20 per month, with a hosted model service that works right after login, plus higher SUPER / ULTRA tiers. Hosted plans don't mean unlimited model usage.
• Team: Coming soon, with SSO, centralized configuration, and model access and quota controls.
• Enterprise: Contact the vendor. Includes an in-VPC gateway, audit logs and dedicated support.

Platforms and notes

• Clients are available for macOS, Windows, Linux, iOS and Android. The download page is cindy.app/download, and the source code is on GitHub at makecindy/cindy.
• Remote tasks run on your computer, which needs to be online with remote access enabled. Your phone won't automatically move offline tasks to the cloud for execution.
• File access is not a full-disk sync.
• "Local-first" doesn't mean your data never leaves your machine. Model inference sends the necessary context to the model service you choose, and companion cloud sync, remote devices and external plugins also transmit related data.
• The costs and limits of your chosen model service still apply. Local models are currently in Beta, and results depend on the model, hardware and runtime environment.