dwgx/WindsurfAPI

Turn Windsurf / Devin Desktop's 100+ AI models (Claude, GPT, Gemini, DeepSeek, Kimi, GLM, SWE) into OpenAI-, Anthropic- & Gemini-compatible APIs. Zero-dependency self-hosted reverse proxy for Claude Code, Cline & Cursor. 把 Windsurf/Devin 云端 100+ 模型变成三套兼容 API。

What it solves

WindsurfAPI provides a way to access over 100 AI models (including Claude, GPT, Gemini, Grok, and various open-source models) by converting the internal protocols of the Windsurf (formerly Codeium/Devin Desktop) platform into standard, compatible APIs. This allows users to integrate these models into existing AI tools and SDKs that expect OpenAI, Anthropic, or Gemini API formats without needing a direct subscription to each individual provider.

How it works

The project acts as an HTTP proxy server that translates incoming standard API requests (OpenAI, Anthropic, Gemini) into Windsurf's internal gRPC protocol. It communicates with a local Windsurf Language Server (LS) binary, which then forwards requests to the Windsurf cloud servers. To manage stability and access, it maintains an account pool with automatic rotation, rate-limit isolation, and failover mechanisms. It also strips provider-specific identity markers from responses to ensure a clean API experience.

Who it’s for

  • Developers who want to use high-end models (like Claude 3.5/4.5 or GPT-5 variants) in their own applications or scripts.
  • AI Agent users using tools like Claude Code, Cline, Cursor, or Aider who want to use Windsurf's model catalog as their backend.
  • Self-hosters looking for a unified API gateway for multiple LLMs.

Highlights

  • Multi-API Compatibility: Simultaneously supports OpenAI (/v1/chat/completions), Anthropic (/v1/messages), and Gemini (/v1beta/models/*) standards.
  • Extensive Model Support: Access to 100+ models across major providers including Anthropic, OpenAI, Google, and various Chinese LLMs.
  • Zero Dependencies: Built with pure Node.js and no npm runtime dependencies, utilizing custom protobuf implementations and image decoding.
  • Account Management: Includes a web dashboard for OAuth login (Google/GitHub), account pool monitoring, and proxy configuration.
  • Agent-Ready: Specifically designed to pass tool_use and tool_result signals, enabling it to work with agentic CLI tools like Claude Code.
  • Resource Efficient: Implements an LS instance pool with memory guards and idle TTL to prevent OOM (Out of Memory) errors on low-spec VPS.