Rivet

Architecture Overview

Rivet uses a node-graph architecture where each node represents an operation (LLM call, text processing, conditional logic, subgraph, etc.). Graphs are defined visually in the desktop IDE and can be exported as JSON project files. The execution engine processes nodes in dependency order, supporting loops, conditionals, and subgraph invocation. A TypeScript runtime library (rivet-node) allows executing Rivet graphs programmatically in Node.js applications, bridging visual design and code deployment.

When to Use Rivet

• Visual agent prototyping and design
• Debugging complex AI agent logic step by step
• Team collaboration on AI prompt engineering
• Testing and validating AI workflows before deployment
• Building reusable AI graph components

Strengths & Weaknesses

Quick Start

// Use rivet-node to execute Rivet graphs programmatically
import * as Rivet from "@ironclad/rivet-node";

// Load a Rivet project file
const project = await Rivet.loadProjectFromFile("./my-agent.rivet-project");

// Create a processor to run the graph
const processor = Rivet.createProcessor(project, {
  graph: "Main Agent Graph",
  inputs: {
    userMessage: {
      type: "string",
      value: "What are the benefits of AI agents?",
    },
  },
  context: {},
  // Provide model configuration
  openAiKey: process.env.OPENAI_API_KEY,
});

// Run and get results
const result = await processor.run();
console.log(result.output.value);

Features at a Glance

Notable Users

Resources