Promptetheus – Trace, detect, and auto-repair AI agent failures

Promptetheus is an open-source observability framework on GitHub that helps developers debug and fix failures in AI agents — autonomous programs that chain together multiple calls to large language models (LLMs) like GPT-4. As agents grow more complex, they often break in hard-to-trace ways: wrong reasoning, missing context, tool misuse. Promptetheus captures each step (the prompt sent, the model's reply, any tool used), lets developers trace failures to a specific step, and offers an automated repair loop that re-runs or patches the bad step. It fills a gap in the current AI engineering stack, where most debugging tools focus on single LLM calls rather than multi-step agent workflows.