DeepSeek open-sources deepseek-recipe for multi-format API conversion
DeepSeek released deepseek-recipe, an open-source collection of Rust libraries and Python bindings that convert API requests into DeepSeek model prompts and outputs. It supports Messages, Chat Completions, and Responses formats, and is MIT-licensed.
China context
- Original name
- 深度求索
- Outside China
- Open weights · github.com
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Developers outside China can use the MIT-licensed Rust and Python libraries to integrate DeepSeek models into their own API services without negotiating separate agreements.
- For investors
- The open-source release of protocol conversion tooling may reduce friction for enterprises adopting DeepSeek models, potentially increasing demand for DeepSeek's inference services.
DeepSeek published deepseek-recipe on GitHub, providing Rust libraries and Python bindings to uniformly convert API requests in different formats into the Conversation format, encode them into prompts for DeepSeek models, and convert model output into responses. The repository supports text, images, thinking, and client tool calls, and includes encoding for DeepSeek V4 and V4.1 conversations. It does not yet support token probabilities, document content, audio/video input, server tool execution, JSON Schema output constraints, or multiple completions per request. The project is MIT-licensed and includes example servers in Rust and Python.
deepseek-recipe abstracts protocol conversion and prompt encoding for DeepSeek V4 and V4.1, with separate crates for core types, encoding, and image preprocessing. It explicitly excludes server-side tool execution and logprobs, indicating a focus on client-side integration. The repository includes Axum and FastAPI mock inference examples, suggesting it is intended as a reference for backend developers.
Developers integrating DeepSeek models into multi-format API services can use deepseek-recipe to reduce custom conversion code, lowering integration cost. The MIT license permits commercial use without attribution, which may accelerate adoption by enterprises building on DeepSeek.
deepseek-recipe lowers the barrier for developers to connect DeepSeek models to existing API ecosystems, potentially increasing model usage. The open-source MIT license allows companies to embed the conversion logic without licensing fees.
The repository's roadmap lists unsupported features such as logprobs, document content, and server tool execution. A verifiable next signal is whether DeepSeek adds these features in subsequent releases, which would indicate expansion of the library's scope.