Rivet Review 2026

Rivet is an open-source visual programming environment for building AI agents with LLMs, created and maintained by Ironclad. Developers design prompt graphs in a node-based desktop app, save them as version-controllable YAML files, and execute them from TypeScript/Node.js applications via the Rivet Node runtime. The remote debugger lets teams observe a graph running inside a production app in real time — the feature that sets Rivet apart from most agent-building tools.

Rivet was created by Ironclad to solve a specific problem: their team was building sophisticated LLM-powered features for their contract management platform and hitting the limits of purely code-based prompt engineering. Complex prompt chains were hard to visualize, harder to debug, and nearly impossible to review collaboratively. Rivet is the tool they built to fix that, then open-sourced. The desktop editor exposes a node-based canvas where each node is a step in a graph — Text nodes, Prompt nodes, Chat nodes for LLM calls, Code nodes for arbitrary JavaScript, Subgraph nodes for composition, plus control-flow nodes for branching, looping, and match/switch logic. LLM providers supported include OpenAI, Anthropic, Google, Anthropic's Claude via Bedrock, and any OpenAI-compatible endpoint (Ollama, LM Studio, vLLM, custom gateways). Graphs are stored as YAML: this makes them diff-friendly in Git, PR-reviewable, and portable across environments. The Rivet Node runtime (@ironclad/rivet-node on npm) loads a project YAML file and executes it programmatically inside a Node.js application. Together with the remote debugger — which uses websockets to stream graph execution back into the desktop editor — teams get a full loop from design to production observability. Rivet supports plugin extensibility: custom nodes, custom LLM providers, and custom plugins can be written in TypeScript and shared as npm packages. The project is MIT-licensed and actively maintained by Ironclad's engineering team, with regular releases on GitHub (run-llama... actually the repo is Ironclad-CX/rivet). Companies publicly using Rivet include Ironclad (contract AI), Attentive (marketing agents), AssemblyAI (transcription pipelines), and Willow Servicing. There is no hosted or commercial version — Rivet is free forever and self-hosted. Best positioned against: LangGraph (Python-first, more community), Flowise (broader no-code, less TypeScript-native), Langflow (Python + hosted option), and CrewAI (agent-team orchestration, Python-first). Rivet is the strongest fit when your team is TypeScript-shaped, needs a shared visual surface for prompt engineering, and wants strong production observability.