Rivet

Ironclad built Rivet so teams could design LLM workflows as node graphs instead of long stretches of glue code. The desktop editor, a Tauri app with installers for macOS, Windows, and Linux, lets you wire chat, text, loop, and subgraph nodes together and then inspect the value flowing through every connection while a graph executes. Finished projects are plain files that your own TypeScript or Node.js service loads and calls like a function through the @ironclad/rivet-core and @ironclad/rivet-node npm packages, and the rivet CLI can execute one graph from the terminal or expose a whole project over a local HTTP server, which is also packaged as a Docker image. Bundled plugins connect to OpenAI, Anthropic, Google, Hugging Face, Pinecone, AssemblyAI, Autoevals, and Gentrace, and the Trivet library adds test suites for graphs. Calls go straight to the providers with your own API keys. The MIT license allows commercial embedding. Commits over the past year are mostly dependency bumps, so expect a stable tool rather than rapid feature growth. Best for TypeScript teams that want visual debugging of agent logic.