DeepClaude

Why we included this project

This self-hosted service pairs a reasoning model like DeepSeek R1 with a generation model, sending the thinking step to DeepSeek and the completion to Claude or Gemini through a single OpenAI-compatible API. The pairing is sometimes described as architect and editor, and the point here is that you get that combination without writing and maintaining your own proxy. It runs in Docker and takes configuration through a web interface, and it speaks the standard OpenAI format, so existing clients like Cherry Studio, NextChat, LobeChat, or Dify can point at it directly. Streaming and non-streaming calls both work, and third-party relay providers can serve as backends, which helps when you don't have direct API access to every model. The practical payoff is swapping model combinations behind one familiar endpoint instead of managing several paid services separately.