Runsight: YAML-First Workflow Engine for AI Agents
Runsight is an open-source, self-hosted workflow engine designed specifically for AI agent orchestration. It takes a unique YAML-first approach where every workflow is defined as a YAML file that lives in your Git repository, making agent workflows as version-controlled and collaborative as traditional code.
Key Features
- Dual Interface: Edit workflows in either a visual canvas or YAML editor with real-time synchronization between both views
- Git-Native Workflows: Every workflow change is a Git commit, enabling full version control, branching, and collaboration
- Cost Tracking: Real-time per-run cost monitoring with hard budget caps to prevent overspending
- Built-in Evaluation: Assertions and regression testing framework for every block output
- Runtime Control: Pause, inspect, and kill running agents mid-execution
- Self-Hosted: Runs entirely on your infrastructure with your API keys and models
How It Works
Runsight operates through a simple command-line interface (uvx runsight) that starts a local server. Users can then design workflows using either the visual canvas or YAML editor, add their API keys, and execute workflows while monitoring costs and performance in real-time. The platform supports complex workflow patterns including loops, sub-workflows, and conditional branching.
