Rivet

Rivet is an open-source visual programming environment for building, debugging, and deploying AI agent pipelines. Developed by Ironclad — the digital contracting platform — Rivet was born out of internal frustration with building complex LLM workflows programmatically. The team open-sourced it after using it to ship their own Contract AI product, which powers a virtual contract assistant across Ironclad's platform.

At its core, Rivet lets developers construct "prompt graphs" — visual node-based pipelines where LLM calls, data transformations, conditionals, and external tool invocations are wired together as connected nodes rather than code. This approach makes complex multi-step agent logic visible and navigable in a way that code alone does not provide.

Rivet's primary audience is engineering teams building production AI features, not just prototypes. The tool distinguishes itself from notebook-based environments (like Jupyter) or purely code-driven orchestration libraries (like LangChain or LlamaIndex) by offering a persistent visual canvas that serves as both the development interface and the documentation. Teams at Attentive, Willow Servicing, and Sourcegraph have cited this visual clarity as meaningful when iterating on agentic logic.

A standout capability is remote debugging: Rivet can observe the execution of prompt chains running inside a live application in real time, not just in a sandbox. This closes the gap between the development environment and production behavior, which is a common pain point when prompt chains behave differently under real data conditions.

Collaboration is handled practically: Rivet graphs are stored as YAML files, meaning they live in version control alongside application code. Teams can diff, review, and merge graph changes using the same pull request workflows they already use. This is a concrete advantage over tools that store pipeline state in proprietary cloud databases.

For deployment, Rivet graphs are designed to run inside Node.js or TypeScript applications via an integration API. The workflow is: design and iterate in the Rivet desktop app, then execute the same graph directly from application code. This avoids the common problem of maintaining separate "prototype" and "production" versions of a pipeline.

Compared to alternatives like Flowise or LangFlow, Rivet is more tightly coupled to a deployment model (Node/TypeScript), which limits flexibility for Python-heavy stacks but creates a more disciplined path to production for JavaScript ecosystems. Against code-only frameworks like LangChain, Rivet offers significantly more visibility into pipeline structure and execution flow at the cost of requiring teams to learn a new visual tool.