Show HN: Record, replay, and improve AI agents in production

Category: infrastructure

Tags: ai-agents, runtime, replay, observability, python, self-hosted

Score: 7.5/10 (Innovation: 7, Technical: 8, Documentation: 8, Utility: 7)

Kitaru is a framework-agnostic runtime for AI agents that records every step of each run as replayable checkpoints, enabling debugging, model swapping, and crash recovery in production. Its innovative combination of checkpoint-based replay with production-grade mechanics like isolated execution and versioned deployments fills a known gap for teams building agent platforms.

Target audience: backend devs, devops, ai-engineers

Repository: https://github.com/zenml-io/kitaru · Python · Apache-2.0 · 197 stars

View on Hacker News