Chinese LLM Models Gateway

OpenClaw native provider plugin for the AIsa unified Chinese model gateway

Install

openclaw plugins install clawhub:@aisa/openclaw-aisa-model-provider

Chinese LLM Models Gateway Plugin for OpenClaw

This package converts the Chinese LLM Models Gateway pull request into a native OpenClaw code plugin that can be published to ClawHub as a Code Plugin.

What this plugin does

The plugin registers a provider named aisa and exposes a unified OpenAI-compatible gateway for major Chinese models through:

AISA_API_KEY
https://api.aisa.one/v1
default model aisa/kimi-k2.5

The bundled catalog currently includes:

Model Ref	Name	Input	Reasoning	Context
`aisa/minimax-m2.1`	MiniMax M2.1	text	No	200k
`aisa/kimi-k2.5`	Kimi K2.5	text	Yes	256k
`aisa/qwen3-max`	Qwen3 Max	text, image	Yes	256k
`aisa/glm-5`	GLM-5	text, image	Yes	200k
`aisa/deepseek-v3.2`	DeepSeek V3.2	text	Yes	128k
`aisa/seed-1-8-251228`	Seed 1.8	text	Yes	128k

Directory layout

text

openclaw-aisa-model-provider/
├── package.json
├── openclaw.plugin.json
├── index.ts
├── README.md
└── src/
    ├── constants.ts
    ├── onboard.ts
    └── provider-catalog.ts

Local installation

bash

openclaw plugins install ./openclaw-aisa-model-provider
openclaw gateway restart

Then configure credentials either by environment variable:

bash

export AISA_API_KEY="your-key"

or through onboarding:

bash

openclaw onboard --auth-choice aisa-api-key

Publish to ClawHub

On ClawHub, upload this folder as a Code Plugin. The platform will inspect the package, detect openclaw.plugin.json, and prefill package metadata.

Notes

This package intentionally extracts the provider behavior from the PR into plugin space. It does not yet include the follow-up reliability fixes mentioned in the PR, such as JSON fence stripping, stricter structured-output prompting, or automatic retry logic. Those should be added either as provider runtime wrappers or as separate companion plugins.