Runsight: YAML-First Workflow Engine for AI Agents
Runsight is an open-source, self-hosted workflow engine designed specifically for AI agent development and orchestration. It introduces a unique "duality" approach where workflows can be designed either visually through a canvas interface or directly as YAML files, with both views staying perfectly synchronized.
Key Features
- YAML-First Workflow Design: Define AI agent workflows as simple YAML files that live in your codebase
- Git-Native Versioning: Every workflow change is tracked through Git commits, enabling proper version control and collaboration
- Real-Time Cost Tracking: Monitor execution costs per block and per run with hard budget caps to prevent overspending
- Built-in Evaluation Framework: Add assertions and validation checks directly within workflow definitions
- Canvas + Editor Duality: Visual canvas for designing workflows alongside a Monaco editor for direct YAML editing
- Runtime Control: Pause, resume, or kill running agents mid-execution
- Self-Hosted Architecture: Runs entirely on your infrastructure with your API keys and models
How It Works
- Installation: Run
uvx runsightto start the local server - Design: Create workflows in YAML or using the visual canvas
- Execute: Run workflows with real-time monitoring of costs and execution
- Version: Commit workflow changes to Git like any other code
Target Users
- AI developers building complex agent workflows
- Teams needing reproducible AI pipelines
- Organizations requiring cost control and audit trails for AI operations
- Developers who want to treat AI workflows as infrastructure-as-code
Unique Value Proposition
Runsight solves the "debugging in the dark" problem common in AI agent development by providing full visibility into workflow execution, cost tracking, and version control. Unlike proprietary platforms, it ensures no vendor lock-in since workflows are standard YAML files that remain usable even if the tool disappears.
