@aid-on/unillm-vercel-ai-sdk

日本語 | English

Purpose

This package wraps the edge-native @aid-on/unillm library to provide Vercel AI SDK compatibility. Use this when:

• Migrating existing AI SDK code to edge-native architecture
• Need LanguageModelV1 interface for compatibility with AI SDK ecosystem
• Want edge optimization without rewriting all your AI SDK calls

For new projects, consider using @aid-on/unillm directly for better edge performance.

Installation

npm install @aid-on/unillm-vercel-ai-sdk

Usage

Drop-in replacement for AI SDK providers:

// Before (using AI SDK providers directly)
import { createGroq } from "@ai-sdk/groq";
import { generateText } from "ai";

const groq = createGroq({ apiKey: process.env.GROQ_API_KEY });
const model = groq("llama-3.1-8b-instant");

// After (using unillm wrapper)
import { getModel } from "@aid-on/unillm-vercel-ai-sdk";
import { generateText } from "ai";

const model = getModel("groq:llama-3.1-8b-instant", {
  groqApiKey: process.env.GROQ_API_KEY,
});

// Same AI SDK calls work unchanged
const result = await generateText({
  model,
  prompt: "Hello!",
});

Multi-Provider Support

import { getModel, createFallbackChain } from "@aid-on/unillm-vercel-ai-sdk";
import { generateText } from "ai";

// Unified interface for all providers
const groqModel = getModel("groq:llama-3.1-8b-instant", credentials);
const geminiModel = getModel("gemini:gemini-2.0-flash", credentials);
const cloudflareModel = getModel("cloudflare:@cf/meta/llama-3.3-70b-instruct-fp8-fast", credentials);

// Fallback chain
const chain = createFallbackChain({
  models: [
    "groq:llama-3.1-8b-instant",
    "gemini:gemini-2.0-flash",
  ],
  credentials,
});

const result = await chain.generate(async (model) => {
  return generateText({ model, prompt: "Hello" });
});

Supported Models

All models from @aid-on/unillm:

Groq

• groq:llama-3.1-8b-instant
• groq:llama-3.3-70b-versatile
• groq:openai/gpt-oss-120b
• groq:openai/gpt-oss-20b

Gemini

• gemini:gemini-2.0-flash
• gemini:gemini-2.0-flash-exp

Cloudflare Workers AI

• cloudflare:@cf/meta/llama-3.3-70b-instruct-fp8-fast
• cloudflare:@cf/meta/llama-3.1-8b-instruct
• cloudflare:@cf/openai/gpt-oss-120b
• cloudflare:@cf/openai/gpt-oss-20b

Migration Guide

From Individual AI SDK Providers

// Before
import { createGroq } from "@ai-sdk/groq";
import { createGoogleGenerativeAI } from "@ai-sdk/google";

const groq = createGroq({ apiKey: "..." });
const google = createGoogleGenerativeAI({ apiKey: "..." });

// After
import { getModel } from "@aid-on/unillm-vercel-ai-sdk";

const groqModel = getModel("groq:llama-3.1-8b-instant", { groqApiKey: "..." });
const geminiModel = getModel("gemini:gemini-2.0-flash", { geminiApiKey: "..." });

Benefits of Migration

• Unified interface across all providers
• Edge-optimized WebStreams support
• Memory efficient for edge environments
• Built-in retry logic with exponential backoff
• Structured output with Zod validation
• Consistent error handling across providers

Advanced Features

All advanced features from @aid-on/unillm are available:

import {
  generateObject,
  withRetry,
} from "@aid-on/unillm-vercel-ai-sdk";

// Structured output
const result = await generateObject({
  model: "groq:llama-3.1-8b-instant",
  credentials,
  schema: z.object({ name: z.string() }),
  prompt: "Generate a person",
});

// Retry logic
const retryResult = await withRetry(
  () => generateText({ model, prompt }),
  { maxAttempts: 3 }
);

API Reference

Wrapper Functions

| Export | Description | |--------|------------| | getModel(spec, credentials) | Create a LanguageModelV1 instance from a model spec | | createFallbackChain(options) | Create a fallback chain across multiple providers |

Re-exported from @aid-on/unillm

| Export | Description | |--------|------------| | parseModelSpec | Parse a model spec string into provider and model | | createModelSpec | Create a model spec from provider and model | | hasCredentials | Check if credentials are available for a provider | | getCredentialsFromEnv | Get credentials from environment variables | | getModelInfo | Get metadata for a model | | getModelsByProvider | List models for a provider | | getAllSpecs | Get all available model specs | | getRecommendedModels | Get recommended models | | isValidSpec | Validate a model spec string | | generateObject | Generate structured output with schema validation | | extractJSON | Extract JSON from LLM response text | | withRetry | Retry wrapper with exponential backoff | | withRetryResult | Retry wrapper returning detailed result | | createRetryWrapper | Create a reusable retry wrapper | | LLMProviderError | Error class for LLM provider errors | | wrapError | Wrap an error as LLMProviderError | | isLLMError | Type guard for LLMProviderError | | isRetryable | Check if an error is retryable | | isRetryableCode | Check if an error code is retryable |

Performance Comparison

| Feature | @aid-on/unillm | @aid-on/unillm-vercel-ai-sdk | |---------|----------------|------------------------------| | Bundle size | ~50KB | ~200KB+ (includes AI SDK) | | Cold start | ~10ms | ~50ms+ | | Memory usage | Minimal | Higher (AI SDK overhead) | | Edge optimization | Native | Compatible | | AI SDK compatibility | No | Full |

Related Packages

• @aid-on/unillm - Edge-native core library (no AI SDK dependency)
• @aid-on/unillm-vercel-ai-sdk - AI SDK compatibility wrapper (this package)

License

MIT