AI Agent Types

Two teams, one agent system. The client team is writing a React app. The server team is writing a Node service that drives the actual agent loop. They both need to agree on what a "message" looks like, what a "tool call" looks like, what shape comes back when the agent finishes thinking. Without shared types, that agreement is a Slack thread and a prayer.

ai-agent-types is a TypeScript type-definitions package for teams building distributed AI agent systems with multiplayer-style clients and services. It does one narrow thing: it gives the client and the server the same vocabulary, in .d.ts form, so TypeScript yells at you when they drift.

This is the kind of library that sounds boring until you've debugged a production incident caused by a field being named toolCalls on one side and tool_calls on the other. Then it sounds like a gift.

If you're the sort of team that keeps client and server in a monorepo, or publishes a shared package for things like API response shapes, this slots in next to that. If your agent system has one frontend, one backend, and a clean boundary between them, you probably don't need a separate types package — but if you have multiple clients (web, mobile, CLI, a voice front-end) all talking to the same agent service, types-as-a-package starts paying for itself fast.

It pairs naturally with the libraries that sit on either side of it — AI Agent Messenger (React) on the UI side, Telnyx AI Agent Library on the telecom side, and whatever agent framework lives in the middle. The types package is the glue the rest of them wire into.

Add it, import it in both client and server, and let the compiler do the nagging.