multi-ai-client

A powerful, production-ready unified AI SDK for JavaScript & TypeScript.

Interact with multiple AI providers using one clean API:

OpenAI · Claude (Anthropic) · Gemini · Mistral

Build AI applications faster with:

✅ Unified API
✅ chat() + stream()
✅ Real-time streaming (AsyncIterator)
✅ Smart retry system
✅ Rate limit handling
✅ TypeScript support
✅ Extensible architecture

✨ Why multi-ai-client?

Every AI provider has a different API.

Different:

• request formats
• authentication methods
• streaming implementations
• response formats
• SDK logic

That means:

❌ more complexity
❌ duplicated code
❌ harder maintenance

multi-ai-client solves this problem by exposing one unified API for every provider.

Write your code once.

Switch providers anytime.

🚀 Features

🤖 Unified API

One interface for:

• OpenAI
• Claude (Anthropic)
• Gemini
• Mistral

No provider-specific code.

💬 Unified Chat API

Simple and consistent API across providers.

await ai.chat("Explain quantum computing")

⚡ Real-Time Streaming

Built-in streaming with AsyncIterator.

for await (const chunk of ai.stream(prompt)) {
  console.log(chunk.content)
}

🔄 Smart Retry System

Automatically retries failed requests.

Supports:

• 429
• 500
• 502
• 503
• 504

Built-in exponential backoff for reliability.

🚦 Rate Limit Handling

Automatically detects:

retry-after

No manual retry logic required.

🧠 TypeScript First

Fully typed API.

Autocomplete support included.

📦 Lightweight

Fast and minimal architecture.

🔌 Extensible Architecture

Designed for future extensibility:

• plugins
• custom providers
• middleware
• caching
• templates

📦 Installation

npm install multi-ai-client

⚡ Quick Start

import { AIClient } from "multi-ai-client"

const ai = new AIClient({
  apiKey: process.env.OPENAI_API_KEY!,
  provider: "openai"
})

const response =
  await ai.chat(
    "Explain artificial intelligence simply"
  )

console.log(response)

🔌 Supported Providers

OpenAI

new AIClient({
  apiKey: "YOUR_KEY",
  provider: "openai"
})

Recommended models

• gpt-4o
• gpt-4o-mini
• gpt-4-turbo

Claude (Anthropic)

new AIClient({
  apiKey: "YOUR_KEY",
  provider: "anthropic"
})

Recommended models

• claude-3-5-sonnet
• claude-3-opus

Gemini

new AIClient({
  apiKey: "YOUR_KEY",
  provider: "gemini"
})

Recommended models

• gemini-1.5-pro
• gemini-1.5-flash

Mistral

new AIClient({
  apiKey: "YOUR_KEY",
  provider: "mistral"
})

Recommended models

• mistral-large-latest
• open-mistral-7b

💬 Chat Example

import { AIClient } from "multi-ai-client"

const ai = new AIClient({
  apiKey: process.env.OPENAI_KEY!,
  provider: "openai"
})

const response =
  await ai.chat(
    "Write a short story about AI",
    {
      temperature: 0.7,
      maxTokens: 300
    }
  )

console.log(response)

⚡ Streaming Example

Real-time AI responses using AsyncIterator.

import { AIClient } from "multi-ai-client"

const ai = new AIClient({
  apiKey: process.env.OPENAI_KEY!,
  provider: "openai"
})

for await (
  const chunk of ai.stream(
    "Tell me a story about space"
  )
) {
  process.stdout.write(
    chunk.content
  )
}

🧠 Stream Output Format

Each chunk follows a unified format across all providers.

{
  content: "Hello",
  done: false
}

Final chunk:

{
  content: "",
  done: true
}

🔄 Switching Providers

Change providers instantly without rewriting your application logic.

OpenAI

provider: "openai"

Claude

provider: "anthropic"

Gemini

provider: "gemini"

Mistral

provider: "mistral"

Example:

const ai = new AIClient({
  apiKey: process.env.API_KEY!,
  provider: "anthropic"
})

const response =
  await ai.chat(
    "Explain black holes"
  )

console.log(response)

⚙️ Chat Options

Customize your AI responses.

await ai.chat(
  "Explain AI",
  {
    model: "gpt-4o-mini",

    temperature: 0.7,

    maxTokens: 500,

    retry: {
      attempts: 5,
      delay: 1000
    }
  }
)

Available Options

| Option | Type | Description | |--------|------|-------------| | model | string | Model to use | | temperature | number | Creativity level | | maxTokens | number | Maximum output tokens | | retry | object | Smart retry configuration |

🔄 Smart Retry Example

Production-ready retry system.

const response =
  await ai.chat(
    "Explain React",
    {
      retry: {
        attempts: 3,
        delay: 1000
      }
    }
  )

Automatically handles:

• temporary provider failures
• network instability
• API overload

Includes:

✅ exponential backoff
✅ automatic retries
✅ provider recovery

🚦 Rate Limit Handling Example

When a provider returns:

429 Too Many Requests

multi-ai-client automatically reads:

retry-after

and retries safely.

No extra logic required.

🏗 Architecture

AIClient
 ├── OpenAI
 ├── Claude (Anthropic)
 ├── Gemini
 └── Mistral

Each provider is internally isolated while exposing:

chat()
stream()

through one consistent API.

🔮 Roadmap

Completed

• ✅ Unified API
• ✅ Multi-provider support
• ✅ chat()
• ✅ stream()
• ✅ AsyncIterator streaming
• ✅ Smart Retry System
• ✅ Rate Limiting Handling
• ✅ TypeScript support

Coming Soon

• 🧩 Prompt Templates
• 🔌 Plugin System for Custom Providers
• ⚡ Token-Level Streaming Parsing
• 💾 Intelligent Cache Layer
• 🌐 Custom API Endpoints
• 🛠 Middleware Support

🛠 Development

npm install
npm run build

📄 License

MIT © Guillaume SERE

⭐ Support

If you like this project:

• ⭐ Star the repository
• 📦 Try the package
• 🐛 Report issues
• 💡 Suggest features

Feedback is always appreciated.