SaladCloud AI SDK Provider

Vercel AI SDK provider for the SaladCloud AI Gateway.

npm install @saladtechnologies-oss/ai-sdk-provider ai@beta zod

Setup

export SALAD_CLOUD_API_KEY="your-api-key"

The provider uses https://ai.salad.cloud/v1 by default. Override it with SALAD_CLOUD_BASE_URL or the baseURL option.

Usage

import { generateText } from 'ai'
import { saladCloud } from '@saladtechnologies-oss/ai-sdk-provider'

const result = await generateText({
  model: saladCloud('qwen3.6-35b-a3b'),
  prompt: 'Say hello in one sentence.',
})

console.log(result.text)

Streaming

import { streamText } from 'ai'
import { saladCloud } from '@saladtechnologies-oss/ai-sdk-provider'

const result = streamText({
  model: saladCloud('qwen3.6-35b-a3b'),
  prompt: 'Say hello in one sentence.',
})

for await (const text of result.textStream) {
  process.stdout.write(text)
}

Custom Configuration

import { createSaladCloud } from '@saladtechnologies-oss/ai-sdk-provider'

const saladCloud = createSaladCloud({
  apiKey: process.env.SALAD_CLOUD_API_KEY,
  baseURL: process.env.SALAD_CLOUD_BASE_URL,
})

Models

Current SaladCloud AI Gateway models:

• qwen3.6-35b-a3b
• qwen3.6-27b
• qwen3.5-9b
• gemma-4-26b-a4b-instruct

Provider Options

Use providerOptions.salad-cloud for SaladCloud-specific options.

For Qwen reasoning models, reasoning: 'none' sends chat_template_kwargs: { enable_thinking: false } to disable thinking. Omitting reasoning, or using low, medium, or high, leaves SaladCloud/vLLM thinking behavior enabled by default.

const result = await generateText({
  model: saladCloud('qwen3.6-35b-a3b'),
  prompt: 'Think briefly, then answer.',
  reasoning: 'low',
  providerOptions: {
    'salad-cloud': {
      topK: 40,
      reasoningEffort: 'medium',
    },
  },
})

Vision

For SaladCloud models that support vision, image URL and inline image data parts from the Vercel AI SDK are sent as OpenAI-compatible image_url content.

import { generateText } from 'ai'
import { saladCloud } from '@saladtechnologies-oss/ai-sdk-provider'

const result = await generateText({
  model: saladCloud('qwen3.6-35b-a3b'),
  messages: [
    {
      role: 'user',
      content: [
        { type: 'text', text: 'Describe this image.' },
        { type: 'file', mediaType: 'image/png', data: { type: 'url', url: new URL('https://example.com/image.png') } },
      ],
    },
  ],
})