Skip to main content
Back

CodePilot Grows From Claude Wrapper to Open-Source Multi-Agent Desktop Powerhouse

CodePilot is a free, open-source desktop AI agent that connects 17+ providers, adds MCP tools, multi-agent orchestration, and phone-based remote control — all in one app.

Repo, Full Access
  • What it is: CodePilot is a free, open-source desktop AI agent client built with Electron + Next.js, supporting 17+ AI providers.
  • Multi-agent orchestration: v0.60.0 added parallel sub-agents with dependency tracking, each assignable to a specific provider and model.
  • Latest models: v0.61.0 added Claude Opus 5, Claude Sonnet 5, and Grok 4.5 Sub-agent support.
  • Extension model: Supports MCP servers (stdio/SSE/HTTP), reusable Skills, and a skills.sh marketplace for prompt patterns.
  • Remote control: Bridge feature lets you control your desktop agent from Telegram, Discord, WeChat, Feishu, or QQ.
  • Free to use: BSL-1.1 license — free for personal/academic use, commercial license required; converts to Apache 2.0 in 2029.

CodePilot started as a GUI wrapper for Claude Code and has since grown into a full multi-model AI agent desktop client. It connects to virtually any AI provider, extends agents with tools via MCP, orchestrates multiple sub-agents in parallel, and lets you control your desktop AI from your phone. With over 6,000 GitHub stars and 109 releases, it is one of the most actively developed open-source AI desktop clients available.

CodePilot main interface showing project chat and assistant dashboard

One client, every model

CodePilot connects to 17+ AI providers out of the box and lets you switch providers and models mid-conversation without losing context. Most AI desktop clients lock you to a single provider or require significant configuration to add others; CodePilot treats provider switching as a first-class feature. The supported roster covers:

  • Western cloud APIs: Anthropic, OpenRouter, AWS Bedrock, Google Vertex AI
  • Chinese providers: Zhipu GLM, Kimi, Moonshot, MiniMax, Volcengine Ark (Doubao), Xiaomi MiMo, Aliyun Bailian (Qwen)
  • Local / self-hosted: Ollama, LiteLLM
  • Custom: Any Anthropic-compatible or OpenAI-compatible endpoint

Beyond chat: a full agent runtime

CodePilot ships three interaction modes: Code, Plan, and Ask, each tuned for a different kind of task. You can set reasoning effort per-session (Low / Medium / High / Max), which maps directly to how much compute the model spends before responding. Sessions can be paused, resumed, and rewound to any checkpoint, which is useful when an agent goes down a wrong path mid-task. The app also includes real-time response streaming, local session persistence, project-aware context with file previews, multimodal attachments, and adjustable permission controls for tool execution.