@meechi-ai/core

The foundational AI & MCP engine for local-first applications.
This library powers the Meechi application, providing the core orchestration layer for Local LLMs, Model Context Protocol (MCP) servers, and Offline-First synchronization.

Features

• Local LLM Orchestration: Abstracted hook (useMeechi) for interacting with local models via WebLLM (WASM).
• MCP Server Protocol: A full TypeScript implementation of the Model Context Protocol for connecting tools and agents.
• Offline-First Storage: Built-in wrappers for IndexedDB (via Dexie) and CRDT sync (via Yjs).
• Hardware Detection: Utilities for detecting GPU capabilities and selecting appropriate quantized models.

Installation

npm install @meechi-ai/core
# or
yarn add @meechi-ai/core

Usage

Basic AI Hook

import { useMeechi } from '@meechi-ai/core';

function App() {
  const { chat, isReady, localAIStatus } = useMeechi();

  const handleSend = async (msg) => {
    await chat(msg, history, context, (chunk) => {
      console.log("Stream:", chunk);
    });
  };

  if (!isReady) return <div>{localAIStatus}</div>;

  return <button onClick={() => handleSend("Hello!")}>Chat</button>;
}

MCP Server Registration

import { mcpRegistry, McpServer } from '@meechi/core';

// Register a custom tool
mcpRegistry.registerTool({
  name: "calculator",
  description: "Perform math",
  handler: async (args) => { return args.a + args.b; }
});

License

This library is available under the AGPLv3 License.