whatbroke-recorder is a Python library that captures detailed traces of AI agent executions, including tool calls and model interactions, saving them in whatbroke JSONL format. It enables developers to diff two runs using the companion whatbroke CLI to precisely identify behavioral changes, regressions, or differences in LLM outputs and agent paths. Designed for AI agent testing, evaluation, and observability, it supports models from providers like OpenAI and Anthropic.
whatbroke-recorder sits in PulseGate's LLM eval & observability category. It focuses on debugging and comparing differences between AI agent runs to identify what changed between executions. It is built as an open-source project for AI developers and researchers. whatbroke-recorder is open source under the MIT license. It runs on the command line and API.
Behind whatbroke-recorder is arthi-arumugam-git, and the product first shipped in 2026. Development happens publicly on GitHub with 34 commits in the last 90 days. Key capabilities include Trace Recording, JSONL Export, and Agent Diffing. It exposes integrations via a public API.
Latest indexed changes and source events
whatbroke-recorder verified by the PulseGate indexer
Other apps tracked under the same category.