DS2API is a Go-based middleware project that converts DeepSeek's web conversation capabilities into APIs compatible with OpenAI, Claude, and Gemini standards. The core backend is implemented in Go, with a React WebUI management console for administration. The project supports multiple deployment methods including local execution, Docker containerization, Vercel serverless deployment, and Linux systemd integration.
The repository's primary purpose is to serve as a reference implementation for high-concurrency protocol adaptation, allowing diverse web protocols to be converted into standardized API formats. The project includes comprehensive API compatibility across multiple platforms, with support for OpenAI's chat completions and responses endpoints, Anthropic's messages interface, Google's Gemini generateContent methods, and Ollama compatibility endpoints. A unified CORS policy is applied across all supported routes to minimize third-party preflight request limitations.
Core capabilities include multi-account token rotation with automatic refresh, supporting both email and phone number login methods. The system implements per-account in-flight request limits with waiting queue management and dynamic concurrent value calculation. DeepSeek's Proof of Work mechanism is implemented in pure Go with millisecond-level response times. Tool calling functionality includes leak prevention through non-code block confidence identification and structured incremental output. The project provides an Admin API for configuration management, runtime settings hot updates, proxy management, account testing, session cleanup, and import/export functionality.
The WebUI management console is a single-page application available at the /admin endpoint, supporting both Chinese and English languages with dark mode support. It enables viewing of server-side conversation records. The system includes operational probes with health check and readiness endpoints. Model support spans multiple DeepSeek variants including deepseek-v4-flash, deepseek-v4-pro, and vision models, with configurable thinking mode and search capabilities. The project maps common model aliases like gpt-4, gpt-5, claude variants, and gemini variants to native DeepSeek models.
According to GitGenius activity tracking, the repository has processed 176 issues and pull requests with a median response latency of 0.0 hours and mean latency of 0.6 hours, indicating active maintenance. Bug reports represent the most active issue label with 93 items, followed by enhancement requests with 51 items. The primary contributor CJackHwang has logged 376 events, with secondary contributors TesseractLHY and shenyaocan contributing 32 and 29 events respectively. The project shares contributors with notable repositories including llvm/llvm-project, lobehub/lobehub, and langgenius/dify, suggesting integration within a broader ecosystem of AI and development tools.
The repository includes comprehensive documentation covering architecture details, API specifications, and configuration guidance. Deployment options range from downloading pre-built release packages to Docker deployment, Vercel serverless functions, and local source code execution. Configuration is centralized through a config.json file, and the project supports both streaming and non-streaming API responses. The implementation includes platform compatibility matrices showing P0 support for Codex CLI/SDK, OpenAI SDK, Vercel AI SDK, Anthropic SDK, and Google Gemini SDK, with P1 support for LangChain, LlamaIndex, and OpenWebUI through OpenAI-compatible interfaces.
