Looopy AI

A reactive, extensible AI agent framework built with TypeScript and RxJS.

Looopy AI provides a powerful, stream-based architecture for creating advanced AI agents that can handle complex, multi-turn conversations and execute tasks in real-time. It's designed for scalability, observability, and extensibility.

Core Packages

• @looopy-ai/core: The heart of the framework, providing the core Agent class and runLoop function, along with tools for building custom agents.
• @looopy-ai/aws: Integrations for AWS services, including Bedrock for LLM providers and S3 for artifact storage.
• @looopy-ai/examples: A collection of examples to help you get started with the framework.

Getting Started

To install the necessary dependencies, run the following command:

pnpm install

Here's a basic example of how to create and use an agent:

import {
  Agent,
  asyncPrompt,
  InMemoryMessageStore,
  LiteLLMProvider,
  literalPrompt,
  localTools,
  tool,
} from '@looopy-ai/core';
import { z } from 'zod';

// LLM provider that points at your LiteLLM proxy
const llmProvider = new LiteLLMProvider({
  baseUrl: 'http://localhost:4000',
  model: 'gpt-4o-mini',
});

// Optional: add local tools with typed schemas
const tools = localTools([
  tool({
    id: 'echo',
    description: 'Echo text back to the caller',
    schema: z.object({ text: z.string() }),
    handler: ({ text }) => ({ success: true, result: text }),
  }),
]);

const promptPlugin = literalPrompt('You are a helpful assistant.');
// or
const promptPlugin = asyncPrompt(async ({authContext}) => {
// load prompt from langfuse or similar prompt manager
// inject user's name from authContext
return prompt;
});

// Create a new agent
const agent = new Agent({
  agentId: 'my-first-agent',
  contextId: 'demo-session',
  llmProvider,
  messageStore: new InMemoryMessageStore(),
  plugins: [promptPlugin, tools],
});

// Start a conversation
const events$ = await agent.startTurn('Hello, world!');

// Subscribe to events to get the agent's response
events$.subscribe({
  next: (event) => console.log(event),
  error: (err) => console.error(err),
  complete: () => console.log('Turn complete.'),
});

Documentation

For more detailed information about the project, including architecture, API reference, and usage guides, please see our full documentation.

Contributing

We welcome contributions! Please read our contributing guidelines to get started.

License

This project is licensed under the ISC License.