kentcdodds/kody

What it solves

Kody provides a portable "home" for AI agents, allowing them to maintain their own memory, API keys, code, and automations. It solves the problem of agent state being locked into a specific host by making these assets portable across any host that supports the Model Context Protocol (MCP).

How it works

Built on Cloudflare Workers and the Model Context Protocol (MCP), Kody acts as a multi-user platform where each user has an isolated environment containing their assistant's state. It uses a distributed architecture of workers to handle different tasks:

• Origin Worker: Manages the Remix UI, OAuth authentication, and request routing.
• Platform Worker: Handles MCP state, mailboxes, and session management using Durable Objects.
• Runtime Worker: Executes package apps and manages the invoke API.
• Jobs Worker: Manages cron jobs and scheduled tasks.

Instead of a massive list of static tools, Kody focuses on a compact MCP surface with powerful search and code execution flows.

Who it’s for

It is designed for users who want a personal assistant platform that is portable across different AI agent hosts (such as ChatGPT) and developers who want to build and maintain stateful AI agents.

Highlights

• MCP-First Architecture: Designed to be compatible with any MCP-compliant AI agent host.
• Portable State: Keeps memory, keys, and code separate from the agent host.
• Multi-user Isolation: Every signed-in user gets a fully isolated assistant environment.
• Cloudflare Native: Leverages Cloudflare Workers, D1 database, KV, and Durable Objects for high performance and scalability.