AI Agent with Durable Streams (Experimental)

Example project demonstrating how to build an AI agent that communicates through durable streams for reliable message delivery and persistence.

Getting Started

git clone https://github.com/rivet-dev/rivet.git
cd rivet/examples/experimental-durable-streams-ai-agent
npm install
npm run dev

Features

• Durable message delivery: Prompts and responses flow through durable streams, ensuring reliable delivery even if components restart
• Streaming AI responses: AI responses stream in real-time through durable streams to the frontend
• Actor-based processing: AI agent runs as a Rivet Actor with persistent state tracking processed prompts

Prerequisites

• Anthropic API Key (set as ANTHROPIC_API_KEY environment variable)
• (Optional) Durable Streams Test UI for inspecting streams and debugging message flow:
  shCopy

  git clone https://github.com/durable-streams/durable-streams.git
  cd durable-streams/packages/test-ui
  pnpm dev
  

Implementation

The architecture uses two durable streams per conversation:

1. Prompt Stream (/conversations/{id}/prompts): Frontend writes user messages, actor consumes them
2. Response Stream (/conversations/{id}/responses): Actor writes AI response chunks, frontend consumes them

Key implementation details:

• AI Agent Actor (src/backend/registry.ts): Defines the aiAgent actor that consumes prompts from durable streams and writes AI responses back
• Frontend (src/frontend/App.tsx): React chat UI that reads/writes to durable streams
• Streams Server (src/streams-server/server.ts): In-memory durable streams server for development
• Message Types (src/backend/types.ts): TypeScript types for prompt and response messages

Resources

• Durable Streams - Reliable message delivery and persistence
• Architecture Slides - Visual overview of the durable streams architecture

License

MIT