Rivet

Rivet is an open-source visual programming environment for building AI agents and LLM-powered applications — instead of writing prompt-chaining code directly, you construct a graph of connected nodes representing prompts, logic, and actions, and debug the execution visually. It's built and used in production by Ironclad, a digital contracting platform company, inside their own AI features. Who it's for: developers building non-trivial LLM applications — multi-step agents, complex prompt chains with branching logic — who want to see and debug the actual execution flow rather than stepping through code, and teams that want the workflow itself to be reviewable (graphs are stored as YAML, so they work with normal version control). Strengths: being built and battle-tested inside a real production product (Ironclad's own contracting platform) is a meaningful signal of real-world robustness beyond a typical open-source side project. Remote debugging lets you observe prompt execution as it happens, which is often the hardest part of building reliable LLM applications to get visibility into. Weaknesses and limitations: a visual, graph-based workflow tool has a real learning curve of its own, and for a genuinely simple single-prompt use case it's likely more overhead than just writing the API call directly — its value is clearest once a project's prompt logic is complex enough to benefit from visualizing it. Real-world scenario: a team building a multi-step AI agent (retrieve data, reason over it, decide on an action, execute) wants to see and debug that whole chain visually instead of tracing through nested function calls in code.