Skip to content

Latest commit

 

History

History
122 lines (93 loc) · 3.03 KB

README.md

File metadata and controls

122 lines (93 loc) · 3.03 KB
Raw

EasyRAG

EasyRAG makes it simple to integrate tools and documents into your AI chats with ease. It's the typescript AI library you have been looking for!

Getting Started

npm install --save easyrag
import EasyRAG, { OpenAIModelAdapter, Tool, Model } from 'easyrag';

// 1. Setup the model adapter, model, and any tools.

const modelAdapter = new OpenAIModelAdapter({
  apiKey: process.env.OPENAI_API_KEY
});

const chatGpt35Turbo = new Model("gpt-3.5-turbo", "chat");

const weatherToolParams = [
  {
    "name": "zipCode",
    "description": "Zip code",
    "type": "string",
    "required": true
  }
];

const weatherTool = new Tool(
  "weather",
  "Looks up the weather by zip code",
  weatherToolParams,
  async (params) => {
    return "It's a crisp 30f in " + params.zipCode;
  }
);

// 2. Initalize the EasyRAG client.

const ragClient = new EasyRAG({ modelAdapter });

// 3. Register your models and tools

ragClient.register(chatGpt35Turbo);
ragClient.register(weatherTool);

// 4. Query the client
await ragClient.query("What is the weather in zip 92021?");

// 5. Get the built-in conversation history
console.log(
  ragClient
    .conversation
    .getMessages()
    .map(m => `${JSON.stringify(m, null, 2)}`)
    .join('\n')
);

Documentation

Documentation is currently under development. Check back soon for updates.

Why & Ethos

I created this library as an alternative to existing tools due to increasing frustration with them. EasyRAG aims to:

  • Provide a minimal API abstraction to easily build a variety of AI-powered applications.
  • Be extendable out-of-the-box. Create your own adapters and integrate with any inference engine (Ollama, LlamaIndex, etc.)
  • Throw errors instead of falling back to defaults.

Feature Roadmap

I am committed to releasing the following features:

  • Model Adapter
    • Support for parameters (temperature, top_p, etc.)
    • Chat completion
    • Embedding
    • Streaming
  • Conversation
    • Hot swap conversation
    • Hot swap tools
    • Hot swap models
    • System prompt
    • History limit for generation context
    • Context - define variables for the AI to use as a "memory"
  • Tool Support
    • Type support for parameters
  • Document Storage Adapter
    • Search nearest K
  • Out-of-the-box Adapters
    • OpenAI Model Adapter
      • Chat
      • Embeddings
      • Tool Support
      • Images (base64)
    • Ollama Model Adapter
      • Chat
      • Embeddings
      • Images (base64)
      • Tool Support
    • PGVector Document Storage Adapter
  • Improved Documentation
    • Official docs/wiki
    • Documented code with descriptions + hints
  • Better testing

Ideas

  • Assign the model to the adapter instead of the client.
  • Swap adapter in query/embedding options

Contributing

Contributions and suggestions for improvement are welcome! Feel free to open an issue or pull request.

License

This project is licensed under the MIT License - see the LICENSE file for details.