@auto-engineer/ai-gateway

Unified AI provider abstraction layer with multi-provider support and MCP tool integration.

Purpose

Without @auto-engineer/ai-gateway, you would have to manage multiple AI provider SDKs, handle provider-specific authentication, and implement your own streaming and structured data generation logic.

This package provides a unified interface for OpenAI, Anthropic, Google, XAI, and custom OpenAI-compatible providers. Built on the Vercel AI SDK with support for text generation, structured data, streaming, and MCP tools.

Installation

pnpm add @auto-engineer/ai-gateway

Quick Start

import { generateTextWithAI, AIProvider } from '@auto-engineer/ai-gateway';

const text = await generateTextWithAI('Explain quantum computing', {
  provider: AIProvider.Anthropic,
  temperature: 0.7,
  maxTokens: 1000,
});

console.log(text);
// → "Quantum computing is a type of computation..."

How-to Guides

Generate Text

import { generateTextWithAI } from '@auto-engineer/ai-gateway';

const text = await generateTextWithAI('Write a haiku about coding');

Stream Text

import { streamTextWithAI } from '@auto-engineer/ai-gateway';

for await (const chunk of streamTextWithAI('Tell me a story')) {
  process.stdout.write(chunk);
}

Generate Structured Data

import { generateStructuredDataWithAI, z } from '@auto-engineer/ai-gateway';

const TodoSchema = z.object({
  title: z.string(),
  priority: z.enum(['low', 'medium', 'high']),
  completed: z.boolean(),
});

const todo = await generateStructuredDataWithAI('Create a todo for code review', {
  schema: TodoSchema,
  schemaName: 'Todo',
});

Use with Images

import { generateTextWithImageAI, AIProvider } from '@auto-engineer/ai-gateway';

const description = await generateTextWithImageAI(
  'Describe this image',
  imageBase64String,
  { provider: AIProvider.OpenAI }
);

Register MCP Tools

import { registerTool, startServer, z } from '@auto-engineer/ai-gateway';

registerTool<{ name: string }>(
  'greet',
  {
    title: 'Greeting Tool',
    description: 'Greets users',
    inputSchema: { name: z.string() },
  },
  async ({ name }) => ({
    content: [{ type: 'text', text: `Hello, ${name}!` }],
  })
);

await startServer();

API Reference

Package Exports

import {
  generateTextWithAI,
  generateTextStreamingWithAI,
  streamTextWithAI,
  generateTextWithImageAI,
  generateTextWithToolsAI,
  generateStructuredDataWithAI,
  streamStructuredDataWithAI,
  getAvailableProviders,
  getDefaultAIProvider,
  getDefaultModel,
  AIProvider,
  z,
} from '@auto-engineer/ai-gateway';

import { createAIContext, generateText } from '@auto-engineer/ai-gateway/core';

import { registerTool, startServer, mcpServer } from '@auto-engineer/ai-gateway/node';

Entry Points

| Entry Point | Import Path | Description | |-------------|-------------|-------------| | Main | @auto-engineer/ai-gateway | Node.js wrappers with global context | | Core | @auto-engineer/ai-gateway/core | Pure functions requiring explicit context | | Node | @auto-engineer/ai-gateway/node | Full Node.js API including MCP server |

Functions

generateTextWithAI(prompt: string, options?: AIOptions): Promise<string>

Generate text from a prompt.

streamTextWithAI(prompt: string, options?: AIOptions): AsyncGenerator<string>

Stream text generation as an async generator.

generateStructuredDataWithAI<T>(prompt: string, options: StructuredAIOptions<T>): Promise<T>

Generate structured data validated against a Zod schema.

AIProvider

enum AIProvider {
  OpenAI = 'openai',
  Anthropic = 'anthropic',
  Google = 'google',
  XAI = 'xai',
  Custom = 'custom',
}

AIOptions

interface AIOptions {
  provider?: AIProvider;
  model?: string;
  temperature?: number;
  maxTokens?: number;
  streamCallback?: (token: string) => void;
  includeTools?: boolean;
}

Architecture

src/
├── index.ts
├── core/
│   ├── context.ts
│   ├── generators.ts
│   ├── types.ts
│   └── providers/
└── node/
    ├── wrappers.ts
    ├── config.ts
    └── mcp-server.ts

The following diagram shows the provider abstraction:

flowchart LR
    A[Application] --> B[ai-gateway]
    B --> C[OpenAI]
    B --> D[Anthropic]
    B --> E[Google]
    B --> F[XAI]
    B --> G[Custom]

Flow: Application calls ai-gateway, which routes to the configured provider.

Dependencies

| Package | Usage | |---------|-------| | ai | Vercel AI SDK core | | @ai-sdk/anthropic | Anthropic provider | | @ai-sdk/openai | OpenAI provider | | @ai-sdk/google | Google Gemini provider | | @ai-sdk/xai | xAI Grok provider | | zod | Schema validation |