Show HN: Record, replay, and improve AI agents in production (github.com)

Kitaru has announced a new self-hosted, framework-agnostic runtime for autonomous agents, designed to enhance the development and deployment processes for AI systems. This tool records every action taken by an AI agent—including model calls, tool calls, and decision points—allowing developers to easily replay executions, diagnose failures, and make informed adjustments. Kitaru integrates seamlessly with existing tech stacks, enabling teams to retain their chosen models and frameworks while adding essential features like recording, replaying, and versioning to their agent workflows. The significance of Kitaru lies in its ability to provide a robust execution layer for AI agents without imposing a specific framework on teams. The system supports features such as crash recovery, isolated execution, and versioned deployments, making it an invaluable resource for organizations looking to streamline their AI operations. With Kitaru, users can pause and resume agent processes, perform detailed comparisons of various runs, and confidently roll out updates, all while maintaining control over their infrastructure. This innovation offers a comprehensive solution for managing the lifecycle of AI agents, fostering reliability and efficiency in production environments.