Skip to main content

​
Get Your Adaptive API Key

Sign up here to create an account and generate your API key.

​
Overview

The Vercel AI SDK works seamlessly with Adaptive through two methods:
  • Adaptive Provider (Recommended): Use the native @adaptive-llm/adaptive-ai-provider provider for built-in support.
  • OpenAI Provider: Use Adaptive via @ai-sdk/openai with a custom base URL.

​
Method 1: Adaptive Provider

​
Installation

npm install ai @adaptive-llm/adaptive-ai-provider

​
Basic Setup

import { adaptive } from "@adaptive-llm/adaptive-ai-provider";

// Use default configuration
const model = adaptive();

​
Method 2: OpenAI Provider

​
Installation

npm install ai @ai-sdk/openai

​
Configuration

import { openai } from '@ai-sdk/openai';
import { generateText } from 'ai';

const adaptiveOpenAI = openai({
  apiKey: process.env.ADAPTIVE_API_KEY,
  baseURL: 'https://api.llmadaptive.uk/v1',
});

const { text } = await generateText({
  model: adaptiveOpenAI(''), // Empty string enables intelligent routing
  prompt: 'Explain quantum computing simply',
});

​
Text Generation

import { generateText } from 'ai';
import { adaptive } from '@adaptive-llm/adaptive-ai-provider';

const { text } = await generateText({
  model: adaptive(),
  prompt: 'Write a vegetarian lasagna recipe for 4 people.',
});

​
Streaming

import { streamText } from 'ai';
import { adaptive } from '@adaptive-llm/adaptive-ai-provider';

const { textStream } = await streamText({
  model: adaptive(),
  prompt: 'Explain machine learning step by step',
});

for await (const delta of textStream) {
  process.stdout.write(delta);
}

​
React Chat Component

import { useChat } from 'ai/react';

export default function Chat() {
  const { messages, input, handleInputChange, handleSubmit } = useChat({
    api: '/api/chat', // Your API route using Adaptive
  });

  return (
    <div>
      {messages.map(m => (
        <div key={m.id}>
          <strong>{m.role}: </strong>
          {m.content}
        </div>
      ))}
      
      <form onSubmit={handleSubmit}>
        <input
          value={input}
          placeholder="Say something..."
          onChange={handleInputChange}
        />
        <button type="submit">Send</button>
      </form>
    </div>
  );
}

​
Configuration Parameters

Advanced configuration options are available with the Adaptive provider for intelligent routing and optimization.
import { generateText } from 'ai';
import { adaptive } from '@adaptive-llm/adaptive-ai-provider';

await generateText({
  model: adaptive(),
  prompt: "Summarize this article",
  providerOptions: {
    adaptive: {
      // Intelligent routing configuration
      model_router: {
        models: [
          "anthropic:claude-sonnet-4-5", // Specific Anthropic model
          "openai:gpt-5-mini" // Specific OpenAI model
        ],
        cost_bias: 0.3, // 0 = cheapest, 1 = best performance
        complexity_threshold: 0.5, // Override complexity detection
        token_threshold: 1000 // Override token threshold
      }
    }
  }
});

​
Parameter Details

Controls intelligent model selection:
  • models: Array of allowed provider:model strings
    • "anthropic:claude-sonnet-4-5" - Specific model
    • "openai:gpt-5-mini" - Specific model
  • cost_bias: Balance cost vs performance (0-1)
    • 0 = Always choose cheapest option
    • 0.5 = Balanced cost and performance
    • 1 = Always choose best performance
  • complexity_threshold: Override automatic complexity detection (0-1)
  • token_threshold: Override automatic token counting threshold
Controls provider fallback behavior:
  • enabled: Enable/disable fallback (default: true)
  • mode: Fallback strategy
    • "sequential" = Try providers one by one (lower cost)
    • "race" = Try multiple providers simultaneously (faster)
Improves performance by caching similar requests:
  • enabled: Enable semantic caching
  • similarity_threshold: Similarity threshold (0-1) for cache hits
    • Higher values = more strict matching
    • Lower values = more cache hits but less accuracy

​
Custom Providers

Custom providers are no longer supported. All models must be specified using the provider:model_name string format for models available in the registry.
Custom Provider Configuration
await generateText({
  model: adaptive(),
  prompt: "Explain machine learning concepts",
  providerOptions: {
    adaptive: {
      // Include registry models in model list
      model_router: {
        models: [
          "openai:gpt-5-mini", // Standard provider
          "anthropic:claude-3-5-haiku"
        ],
        cost_bias: 0.5
      }
    }
  }
});

​
Tool/Function Calling

import { generateText, tool } from 'ai';
import { z } from 'zod';
import { adaptive } from '@adaptive-llm/adaptive-ai-provider';

const { text } = await generateText({
  model: adaptive(),
  prompt: "What's the weather in New York?",
  tools: {
    getWeather: tool({
      description: 'Get weather for a location',
      parameters: z.object({
        location: z.string(),
      }),
      execute: async ({ location }) => {
        return `Weather in ${location} is sunny and 72°F`;
      },
    }),
  },
});

​
Cache Tier Tracking

Access cache information in the response when using the Adaptive provider:
Cache Tier Information
const result = await generateText({
  model: adaptive(),
  prompt: "Hello world",
});

// Check cache tier
console.log(result.usage?.cache_tier);
// "semantic" | undefined

​
Environment Variables

Environment Setup
# .env.local
ADAPTIVE_API_KEY=your-adaptive-api-key

​
What You Get

Intelligent Routing

Automatic model selection based on your prompt complexity

Built-in Streaming

Real-time response streaming with React components

Cost Optimization

Significant cost savings through smart provider selection

Provider Transparency

See which AI provider was used for each request

​
Next Steps

More Examples

See complete working examples

API Reference

Explore all available options and parameters