CodePilot
A multi-model AI agent desktop client — connect any AI provider, extend with MCP & skills, control from your phone. Built with Electron + Next.js.
CodePilot is a desktop application that lets you chat with AI models from many different providers through a single interface. Instead of using a separate website or app for each AI service, you connect your own API keys and switch between providers mid-conversation without losing your place. It runs on macOS, Windows, and Linux.
The app supports over 17 AI providers, including Anthropic, OpenRouter, AWS Bedrock, Google Vertex, several Chinese AI services such as Kimi and Zhihu GLM, and local models running on your own machine through Ollama or LiteLLM. You can also connect any service that uses an OpenAI-compatible or Anthropic-compatible API format, which covers a wide range of self-hosted options.
Beyond basic chat, CodePilot has grown into a broader AI workspace. You can set up a personal profile so the assistant learns your preferences over time through files called soul.md, user.md, and memory.md that you edit yourself. There is also support for MCP servers, which are small plugins that give the AI access to tools like web search, file reading, or database queries. A skills marketplace lets you install reusable task templates. You can run two conversations side by side in a split-screen view, rewind any session back to an earlier point, schedule recurring tasks with timer expressions, and generate images through Google Gemini's image generation API.
A remote bridge feature lets you send messages to your desktop session from Telegram, Discord, WeChat, or similar apps on your phone, and get replies back there. All conversation data is stored locally on your machine using SQLite, so nothing is sent to a central server. Token usage and estimated costs are tracked with daily charts.
One important note: the README states that CodePilot is currently undergoing a significant product refactor before the next release. The existing downloadable versions still work, but the next versions will focus on rebuilding the core session and background task foundations before adding new features.