# zzz-api is four documents pointing at somebody else's endpoint

> zzz-api documents a commercial, OpenAI format API service in Chinese: swap two values, the base URL and the bearer key, and existing OpenAI client code keeps working. The repository itself holds no code, declares neither language nor licence, and publishes no releases.

**xing61/zzz-api** — 企业级稳定的OpenAI、Claude、Gemini等的API接口-For企业和开发者。OpenAI的api proxy，支持ChatGPT的API调用，支持Anthropic claude的官方接口形式，支持Google gemini的官方接口形式，支持：gpt-5，sora。不需要openai Key, 不需要买openai的账号，不需要美元的银行卡，通通不用的，直接调用就行，稳定好用！！智增增

- Repository: https://github.com/xing61/zzz-api
- Website: https://platform.zhizengzeng.com/#/login?aff=github
- Stars: 986 · Forks: 82
- Language: Unknown
- License: not declared
- Published: 2026-09-14 · Updated: 2026-09-14 · Language: en
- Canonical page: https://hysenlabs.com/projects/xing61-zzz-api

## The repository holds documents, not code

The top level of zzz-api contains four entries: README.md, openai-api-desc.md, and two directories, batch/ and fine-tune/. There is no source tree, no package manifest, no container recipe and no test directory, and the project's language and licence are both left undeclared. What you are looking at is a pointer to a hosted service rather than software you can run, since every working instruction ends at a base URL and a key that you collect from somebody else's management console. The two directories match the two features the README keeps returning to, batch requests and fine-tuning, so the repository is a landing page for a commercial endpoint with a little supporting material attached. Nothing in those four entries pins a version of the service or says what happens to a request when the upstream provider changes underneath it.

## The whole integration is two substitutions

Every documented route replaces two values and nothing else. The official base URL, https://api.openai.com/v1/, becomes https://api.zhizengzeng.com/v1/, and the official key, an sk- prefixed string, is replaced by a key issued from the service's management console. The docs state that the endpoint supports https and that all interfaces share the base URL https://api.zhizengzeng.com/. Authentication is a bearer token carried in the request headers, and the header block is given directly:

```
Content-Type: application/json
Authorization: Bearer api_key
```

Because the compatibility target is the OpenAI request and response shape rather than OpenAI itself, code written against that format keeps running unchanged, and that is the real product claim here. The cost sits in the same two substitutions: the key you hold is a credential for this provider, not for the model vendor, so it also carries this provider's pricing, uptime and data handling.

## Three documented ways in, and nothing about errors

Three routes are spelled out, differing only in where the two values go. In Python you use the official openai package, set api_key to the console key and set base_url to https://api.zhizengzeng.com/v1/. In LangChain you set two environment variables instead, OPENAI_API_KEY and OPENAI_API_BASE_URL, with that same base URL. The third route is language agnostic: send HTTP requests yourself, put the key in the header, and keep the official URL shape with only the domain replaced. Two of those are drop-in substitutions for code you already run, and the third is the one to reach for when a client library insists on hardcoded endpoints. What none of the three covers is failure. There are no status codes, no retry advice, no timeout guidance, and no description of what a response looks like when the upstream model rejects a request or when the reseller's quota is exhausted.

## One endpoint claiming four vendor shapes and a long model list

The compatibility claim is broad on both axes. On the vendor side, the docs say the endpoint speaks the official interface form of Anthropic Claude, of Google Gemini and of xAI Grok, and they also name Baidu Wenxin Yiyan, Alibaba, iFlytek Spark and Zhipu ChatGLM alongside ChatGPT. On the OpenAI side the surface named covers video, batch, assistant, fine-tuning, o1-preview, o1-mini, gpt-5.6, gpt-5.5 and sora-2, plus gpt-4o, gpt-4-vision, GPT-3.5, Embedding, Whisper with transcriptions and translations, tts-1 and tts-1-hd, DALL-E-3, function_call and stream mode. Several of those names carry their version inside the string itself, which means the list changes when the README is edited rather than when a release is cut. Read it as a snapshot of what the console offered on the day the sentence was written.

## The homepage is a login page carrying an affiliate code

The project homepage field points at a console login URL with an aff=github query parameter attached, so the front door of the repository is a sign-in page with referral tracking built into it. Inside, the console is described as the place to collect a separate Secret Key, review call records, inspect fine-tuned models, check the balance and copy sample code, and a WeChat group is offered as the support channel. The commercial terms are stated as advantages rather than as a price list: no vendor account, no US dollar bank card, top up by WeChat or pay by company invoice, and a claim of carrying no risk of accounts being banned. No rate limits, no prices, no status page and no statement about whether prompts are logged or reused appear in any of the four entries.

## Chinese only documentation with examples that live elsewhere

Everything a user is told here is written in Chinese, and no English version exists in the tree. The feature list, the three usage routes and the notes section are all Chinese, which fits the audience the endpoint names, and it is the first thing a reader outside that market hits. The openai-api-desc.md file and the batch/ and fine-tune/ directories are the only supporting material, and several feature claims, batch, assistant creation, fine-tuning and function calling among them, end with a pointer to examples instead of to code in this repository. The README does not say whether those examples live here or in the console, and the two directories hold no visible description of what they contain.

## No releases, so version numbers live inside the README text

The repository has no GitHub releases, and there is nothing else to version against: no tags are referenced, no changelog file is present, and the last push here is dated 13 September 2026. Version identity therefore lives in prose, where names like gpt-5.6, gpt-5.5 and sora-2 appear inside a feature list rather than in a manifest that code could be checked against. For a proxy of this kind that difference is practical rather than cosmetic: a client pins a model string and a base URL, and when the service retires a name the only notice is whatever the console or the README says that day. Treat the list as advertising copy rather than a compatibility contract, and confirm in the console before building on any single name in it.

## Conclusion

Read the upstream model vendors' terms of service before you send anything through a reseller like this one, then decide whether a single base URL and a key from a web console is acceptable for your workload. If you take it, verify in the console which model names are live today, since the repository has no releases and the version numbers exist only inside README prose, and treat the Chinese only documentation, the WeChat support channel and the absence of any stated limits or uptime record as part of what you are buying. Teams that need their own credentials, their own audit trail or their own data handling should skip it.

## FAQ

### Is zzz api down?

Nothing in the repository reports service status. The only endpoint named anywhere is the base URL https://api.zhizengzeng.com/, and the console where keys, call records and balance live is platform.zhizengzeng.com, so uptime has to be checked there rather than here.

### What exactly do I change to call OpenAI models through zzz-api?

Two values. Point base_url at https://api.zhizengzeng.com/v1/ instead of https://api.openai.com/v1/, and replace the official sk- prefixed key with the API_KEY issued from the management console, sent as an Authorization: Bearer header.

### Does zzz-api work with LangChain and the official openai package?

Both routes are documented. The Python route uses the official openai package with api_key and base_url set, and the LangChain route sets the OPENAI_API_KEY and OPENAI_API_BASE_URL environment variables instead, pointing at the same base URL.

### Which vendor interfaces can zzz-api speak besides OpenAI?

The docs claim the official interface forms of Anthropic Claude, Google Gemini and xAI Grok, and they also name Baidu Wenxin Yiyan, Alibaba, iFlytek Spark and Zhipu ChatGLM alongside ChatGPT.

### What does zzz-api cost and how are accounts paid for?

No prices or rate limits are published in the repository. What is stated is that you need no vendor account and no US dollar bank card, that top ups can be made by WeChat or by company invoice, and that the service claims no risk of accounts being banned.

## Sources

- [Issues](https://github.com/xing61/zzz-api/issues)
- [Project website](https://platform.zhizengzeng.com/#/login?aff=github)
- [README](https://github.com/xing61/zzz-api/blob/main/README.md)
- [xing61/zzz-api on GitHub](https://github.com/xing61/zzz-api)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/xing61-zzz-api