Vertex Partner Provider

Vercel AI SDK provider for Vertex AI partner models (GLM, Kimi, DeepSeek, MiniMax, Qwen) using the OpenAI-compatible format.

The Problem

Vertex AI partner models use OpenAI-compatible format, but @ai-sdk/google-vertex only supports Gemini format. This provider bridges that gap using the same pattern as @ai-sdk/google-vertex/anthropic.

How It Works

┌─────────────────────────────────────────────────────────────────┐
│                    @ai-sdk/google-vertex                        │
├─────────────────────┬───────────────────────────────────────────┤
│  Main (Gemini)      │  /anthropic submodule                     │
│                     │                                           │
│  Uses:              │  Uses:                                    │
│  GoogleGenerative   │  AnthropicMessagesLanguageModel           │
│  AILanguageModel    │  (from @ai-sdk/anthropic/internal)        │
│                     │                                           │
│  Format: Gemini     │  Format: Anthropic (native)               │
└─────────────────────┴───────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────────────┐
│              vertex-partner-provider (this package)             │
├─────────────────────────────────────────────────────────────────┤
│  Uses:                                                          │
│  OpenAICompatibleChatLanguageModel                              │
│  (from @ai-sdk/openai-compatible)                               │
│                                                                 │
│  Format: OpenAI (native) + Google Cloud Auth                    │
│                                                                 │
│  Endpoint: /endpoints/openapi/chat/completions                  │
└─────────────────────────────────────────────────────────────────┘

Installation

npm install @ai-sdk/openai-compatible google-auth-library ai
# Then copy index.js to your project

Usage

const { createGLM, createDeepSeek, createKimi } = require('./vertex-partner-provider');
const { generateText } = require('ai');

// Create provider for GLM
const glm = createGLM({
  project: 'your-gcp-project',
  location: 'global',  // or specific region
});

// Use it
const result = await generateText({
  model: glm('glm-4.7-maas'),
  prompt: 'Hello!',
});

console.log(result.text);

Available Factory Functions

| Function | Publisher | Models | |----------|-----------|--------| | createGLM() | zai-org | glm-4.7-maas | | createKimi() | moonshotai | kimi-k2-thinking-maas | | createDeepSeek() | deepseek-ai | deepseek-v3.2-maas | | createMiniMax() | minimaxai | minimax-m2-maas | | createQwen() | qwen | qwen3-coder-480b-a35b-instruct-maas |

Generic Usage

const { createVertexPartner } = require('./vertex-partner-provider');

const provider = createVertexPartner({
  project: 'your-project',
  location: 'global',
  publisher: 'any-publisher',
});

const model = provider('model-id');

Authentication

Uses Google Cloud Application Default Credentials:

gcloud auth application-default login
gcloud auth application-default set-quota-project YOUR_PROJECT

Tool Calling Support

⚠️ Note: Tool calling should work with this provider since it uses the native OpenAI-compatible format. The previous issues were caused by using the Gemini format class with OpenAI-format responses.

License

MIT