Runsight — YAML-first workflow engine for AI agents is an AI-powered development platform designed to streamline the creation, management, and deployment of autonomous agent workflows. By treating agent logic as code, it allows developers to define complex operational sequences using YAML files that integrate seamlessly with standard Git version control systems. This approach ensures that every change to an agent's behavior is trackable, reviewable, and reproducible within a team's existing development lifecycle. The platform features a dual-interface system where users can toggle between a visual canvas and a code editor, ensuring that the visual representation and the underlying configuration stay perfectly synchronized. This allows for rapid prototyping of multi-step processes, including loops, conditional gates, and nested sub-workflows. One of its standout capabilities is the granular monitoring system, which provides real-time tracking of execution costs and token usage for every individual block within a run. This is complemented by a robust evaluation framework that uses built-in assertions and transform hooks to validate outputs and maintain high quality across iterations. Built for developers and engineers who prioritize security and control, Runsight is an open-source, self-hosted solution that keeps API keys and model interactions within a local environment. It is particularly well-suited for building research pipelines, automated content generation systems, and complex decision-making agents that require human-in-the-loop oversight. Users can pause or terminate executions mid-flight to prevent resource waste, making it a practical tool for managing production-grade AI agents with precision and transparency.
Problem: Developers fear runaway costs when testing autonomous AI agents.
Solution: Runsight allows setting hard budget caps per run within the YAML config.
Example: An agent automatically kills its own process if the API cost hits $0.20.
Problem: Changes to agent logic are often hidden in messy Python files and hard to review.
Solution: Workflows are defined in YAML and managed via standard Git version control.
Example: A team lead reviews a 'git diff' to see exactly how an agent's logic was updated.
Problem: It is hard to know if an agent update actually improved performance.
Solution: Built-in assertions and regression testing evaluate every block output.
Example: An assertion checks that a research agent's output is always in valid JSON format.
Target audience: Best for: AI Engineers, DevOps Teams, Software Developers
Pricing: Open Source · Categories: Chatbot Development, Developer Tools, Low-code/No-code
Tags: ai agent, developer tools, low-code/no-code, OpenSource, research