[Docs]: how should users generate a custom registry api.json? 用户应该如何生成 custom registry api.json?

Hi, I noticed that Kimi Code supports importing providers from a custom registry:

kimi provider add https://registry.example.com/v1/models/api.json --api-key YOUR_KEY

The docs explain how to import the registry, but I couldn't find much about how users are supposed to create the api.json in the first place.

My use case is connecting a few OpenAI-compatible / Anthropic-compatible endpoints. The model lists can change, so hand-writing the whole registry is a bit awkward. It would be helpful if the docs had a small example of:

• what a minimal api.json looks like
• which fields are required
• how provider type / model capabilities / context size should be represented
• what gets refreshed after importing a registry

Longer term, maybe Kimi Code could also support generating/importing this directly from common model discovery endpoints, like OpenAI-compatible /models.

For now I made a small workaround for myself while trying to solve this:

https://github.com/ayu-exorcist/kimi-registry-adapter

# Try it with one command
npx @kastral/kra

你好，我注意到 Kimi Code 支持从 custom registry 导入 providers：

kimi provider add https://registry.example.com/v1/models/api.json --api-key YOUR_KEY

文档里说明了如何导入 registry，但我没找到太多关于用户应该如何创建这个 api.json 的说明。

我的使用场景是接入几个 OpenAI-compatible / Anthropic-compatible endpoint。模型列表可能会变化，所以手写整个 registry 有点麻烦。如果文档里能有一个小示例，会很有帮助，比如：

• 最小可用的 api.json 长什么样
• 哪些字段是必填的
• provider type / model capabilities / context size 应该如何表示
• 导入 registry 之后，哪些内容会被刷新

长远看，也许 Kimi Code 也可以支持直接从常见的模型发现接口生成或导入，例如 OpenAI-compatible 的 /models。

目前我在解决这个问题时，自己做了一个小的 workaround：

https://github.com/ayu-exorcist/kimi-registry-adapter

# Try it with one command
npx @kastral/kra

Start with the kimi provider add documentation and the custom registry api.json entry point described in the issue. Trace the existing registry format and import refresh behavior, then document a minimal example, required fields, provider and capability values, and context-size representation. Done means users can create and import a changing custom registry without guessing.