Rivet

Rivet is an open-source, node-based visual programming environment for designing complex LLM chains and AI agents. Built and battle-tested by Ironclad for their contract-AI workloads, it ships as a desktop app plus a TypeScript/Node.js runtime, letting you prototype prompt graphs visually then embed them in production code. Graphs are serialized as version-controlled YAML, so teams can review agent logic in pull requests the way they review code.

Where Rivet stands out is the developer workflow: real-time remote debugging lets you attach the IDE to a running Node process and watch node execution live, which is a genuinely rare capability among agent builders. It's model-agnostic (OpenAI, Anthropic, and any provider you wire up), free to use, and aimed at engineering teams who want visual clarity without giving up Git, types, or self-hosting. If you're a solo hacker who prefers plain Python or LangChain, this will feel heavy; if you're a team shipping agent features inside a TypeScript app, it's a strong fit.

Contributors include Ironclad, AssemblyAI, and Sourcegraph, which gives it more real-world provenance than most flowchart-style agent tools. Because it's fully open source under an MIT-style license and there's no hosted control plane, you own the deployment surface entirely — for better and for worse.