Rivet — Visual AI Prompt Workflow IDE

What it is

Rivet is a visual development environment for designing AI prompt workflows. Instead of writing prompt chains in code, you build them by dragging and dropping nodes onto a canvas. Nodes represent LLM calls, conditional branches, loops, data transformers, and external API calls. Connections between nodes define the data flow.

Rivet targets prompt engineers, AI product builders, and teams prototyping complex LLM pipelines. The visual approach makes it easier to reason about multi-step prompt chains and debug failures at each node.

How it saves time or tokens

Debugging prompt chains in code requires adding print statements, re-running the entire chain, and manually inspecting intermediate outputs. Rivet shows the output of every node in real time as the chain executes. You can pause at any node, edit the prompt, and re-run from that point — avoiding wasted tokens on upstream nodes that already succeeded.

How to use

1. Download and install Rivet from the official website or GitHub releases.
2. Create a new project and drag LLM call nodes onto the canvas.
3. Connect nodes with edges to define data flow, then press Run.

# Example node graph (conceptual):
Node 1: Text Input -> 'Summarize this article: {{article}}'
Node 2: LLM Call (GPT-4) -> receives Node 1 output
Node 3: Conditional -> if length > 200, route to Node 4; else Node 5
Node 4: LLM Call -> 'Shorten this summary to 2 sentences'
Node 5: Output -> final summary

Example

// Rivet also provides a Node.js SDK for running graphs programmatically
import { runGraph } from '@ironclad/rivet-node';

const result = await runGraph({
  graphFile: './my-workflow.rivet',
  inputs: {
    article: 'The latest advances in transformer architectures...'
  },
  context: {
    openAiKey: process.env.OPENAI_API_KEY
  }
});

console.log(result.output);

Related on TokRepo

• Automation tools — Explore more visual workflow builders for AI tasks
• Prompt library — Find reusable prompt templates to use in Rivet nodes

Common pitfalls

• Large graphs with many LLM call nodes can become expensive quickly; use mock nodes during development to avoid burning API credits on every test run.
• Rivet stores API keys in the project file by default; use environment variables instead to avoid accidental key exposure when sharing project files.
• Loop nodes without proper exit conditions can run indefinitely and consume tokens until the API rate limit or budget cap is hit.