AI-Generated Actor

Use an AI chat to generate and iterate on Rivet Actor code, then deploy and test it live.

Getting Started

git clone https://github.com/rivet-dev/rivet.git
cd rivet/examples/ai-generated-actor
pnpm install
pnpm dev

Prerequisites

OpenAI API key set as OPENAI_API_KEY
Install dependencies so the secure-exec package is present. This example uses the secure-exec package from pkg.pr.new.
If you need to override the runtime package location, set RIVETKIT_DYNAMIC_SECURE_EXEC_SPECIFIER to a resolvable secure-exec entry file URL.

Features

AI-driven code generation using GPT-4o via the Vercel AI SDK with streaming responses
Dynamic actor loading via dynamicActor from rivetkit/dynamic
Per-key isolation where each actor key has its own AI agent, generated code, and dynamic actor instance
Generic actor interface to call arbitrary actions on the generated actor
Three-column layout: chat, generated code, and actor testing interface

Implementation

The project uses two actors defined in src/actors.ts:

codeAgent maintains chat history and generated code in actor state. It processes messages via a queue and streams AI responses using the Vercel AI SDK, extracting code blocks from the response to update the current actor source.
dynamicRunner is a dynamic actor that loads its source code from the codeAgent with the matching key, executing the AI-generated code in a sandboxed isolate.

The server in src/server.ts exposes proxy endpoints for calling actions on the dynamic actor by name.

The frontend in frontend/App.tsx provides a three-column interface for chatting with the AI, viewing generated code, and testing the deployed actor.

Resources

Read more about dynamic actors, queues, events, and state.

License

MIT