Runsight

Runsight is an open-source, Apache 2.0-licensed workflow engine that lets you design, version, and run AI agent pipelines as YAML files stored in Git. Instead of scattering agent logic across Python scripts, you define workflows in a readable YAML format that lives in your repository—so every change is diffable, reviewable, and revertible like any other code. This makes Runsight a fit for developer teams who treat agent behavior as code and want to collaborate through branch, review, and merge cycles. A key differentiator is the duality between a visual canvas and a Monaco-based YAML editor—both stay bidirectionally synced. Drag nodes on the canvas and the YAML updates; edit YAML and the canvas reacts. Every workflow is a file on disk, and every save is effectively a commit, enabling Git-native version control. Runsight also tracks per-run cost per block, including latency, and lets you set hard budget caps (max_cost and max_tokens) that automatically kill execution before overspend—no surprise bills. The built-in evaluation framework uses assertions on block outputs, with transform hooks for structured extraction and regression testing across runs. You can pause, inspect, and resume or kill agents mid-run. The engine supports loops with break conditions, sub-workflows as child steps, and hierarchical orchestration. Runsight is self-hosted: it runs on your machine with your own API keys and models, and nothing leaves your network unless you choose to send it. Compared to fully-managed agent platforms like Langfuse or agent studios, Runsight requires you to handle your own infrastructure and API keys, but for developer-focused teams already using Git, it's a transparent, lock-in-free alternative that puts your data and config in your hands.