one-api

A key management and redistribution system unifying OpenAI, Azure, Claude, Gemini, DeepSeek, and a long list of Chinese providers under a single API, shipped as one executable file.

One API is an LLM API management and distribution system. Its purpose is to put multiple providers behind a single API so keys can be managed and access redistributed. The supported list is long: OpenAI's ChatGPT series, Anthropic Claude, Google PaLM2 and Gemini, Mistral, ByteDance Doubao, Baidu Ernie, Alibaba Qwen, iFlytek Spark, Zhipu ChatGLM, 360 Zhinao, Tencent Hunyuan, Moonshot AI, and DeepSeek, plus Azure. JavaScript under MIT, around 36,400 stars.

Feature-wise the system handles load balanced access to multiple channels, stream mode for typewriter style output, and multi machine deployment. Token management covers expiry, quota, allowed IP ranges, and allowed models. There is also redemption code management with batch generation and export, and support for image mirroring and third party proxy services.

Docker images come in stable, preview, and alpha variants, with a GitHub mirror for image pulls, and a watchtower based command handles updates. The distribution is a single executable file, and the UI ships in English. The README also carries a compliance note worth reading: users must follow OpenAI usage terms and applicable laws, and must not provide unregistered generative AI services to the Chinese public. The homepage is oneapi.justsong.cn.

Anyone juggling keys across several providers, or running a shared gateway for a team, is the obvious audience. The redistribution features, tokens with quotas and redemption codes, point at scenarios where API access itself is the product being handed out to other people.

A pragmatic piece of plumbing for a fragmented provider market. The compliance note in the README is unusual, and worth taking seriously.

• Source 1 · Project repository
• Source 2 · README