Skip to content

Rivet

What it is

Rivet is an open-source visual AI programming environment and TypeScript library. It allows developers to build, test, and debug complex multi-agent AI systems using a node-based editor. As of May 2026, it has expanded into a comprehensive "Infrastructure for the agentic era" with the launch of agentOS.

What problem it solves

It provides a powerful visual interface for designing AI logic, making it easier to manage complex flows, debug issues, and collaborate on agentic behaviors. With agentOS, it solves the high cost and slow cold-starts of traditional sandboxed environments for agentic code execution.

Where it fits in the stack

Framework / Visual Orchestrator / Agent Runtime.

Typical use cases

  • Agent Design: Crafting intricate logic for autonomous or semi-autonomous AI agents using a node-based editor.
  • Stateful Workflows: Implementing durable, replayable workflows for TypeScript that support sleep, join, and human-in-the-loop.
  • Sandboxed Code Execution: Running untrusted AI-generated code in agentOS (Wasm/V8 isolates) with ~6ms cold starts.
  • Edge Data Storage: Utilizing SQLite for Rivet Actors to manage millions of isolated databases at the edge.

Strengths

  • Developer-Centric: Built for developers with a focus on extensibility and powerful visual debugging.
  • Performance: agentOS provides a full POSIX environment that is 32x cheaper than traditional sandboxes.
  • Durable Execution: Rivet Workflows provide robust state management and replayability.
  • Unified Agent API: The Sandbox Agent SDK provides a single API for integrating with various coding agents like Claude Code or Amp.

Limitations

  • Learning Curve: The visual editor and its various node types can take some time to master.
  • Transitioning Ecosystem: Rapidly evolving from a visual editor into a full-stack agent infrastructure provider.

When to use it

  • When building sophisticated AI agents that require complex logic, state management, and durable workflows.
  • When you need a high-performance, low-cost sandbox for executing AI-generated code.
  • When you value high-quality visual debugging and inspection tools.

When not to use it

  • For trivial AI tasks where a simple API call is sufficient.
  • If you are not comfortable with node-based visual programming for logic design.

Getting started

Installation

To use Rivet in your project: 

npm install @ironclad/rivet-node

Running a Graph

You can load and run a .rivet-project file in your Node.js application:

import { runGraph, loadProject, NodeId } from '@ironclad/rivet-node';

async function runRivetGraph() {
  const project = await loadProject('path/to/project.rivet-project');

  const results = await runGraph(project, {
    graph: 'Main Graph' as NodeId,
    inputs: {
      userInput: { type: 'string', value: 'Hello Rivet!' }
    },
    // Required if using OpenAI nodes
    openAiKey: process.env.OPENAI_API_KEY,
  });

  console.log(results.output.value);
}

runRivetGraph();

Rivet Workflows Example

import { workflow } from '@ironclad/rivet-node';

const myWorkflow = workflow('example', async (c) => {
  const result = await c.callAgent('researcher', { query: 'May 2026 AI trends' });
  await c.sleep('1 day'); // Durable sleep
  return await c.callAgent('writer', { data: result });
});

Licensing and cost

  • Open Source: Yes (MIT License)
  • Cost: Free (OSS version); Usage-based for Rivet Cloud/agentOS.
  • Self-hostable: Yes

Sources / References

Backlog

  • [x] Perform quarterly technical freshness audit. (Completed: 2026-05-31)

Contribution Metadata

  • Last reviewed: 2026-05-31
  • Confidence: high