trajectory-causal-attribution is a Python library that records trajectories of AI agents and uses counterfactual step-ablation replay combined with Shapley value techniques to determine which individual actions led to failures. It helps developers and researchers debug and improve the reliability of LLM-based autonomous agents. The tool is designed for integration into agent frameworks and evaluation pipelines.
trajectory-causal-attribution is a LLM eval & observability product. It focuses on debugging which specific step in an AI agent's trajectory caused an overall failure. It is built as an open-source project for developers. trajectory-causal-attribution is open source under the MIT license. It runs on the command line, and it can be self-hosted.
Behind trajectory-causal-attribution is Krishddd, and the product first shipped in 2026. Development happens publicly on GitHub with 29 commits in the last 90 days. Key capabilities include trajectory recording, counterfactual replay, and step ablation.
Latest indexed changes and source events
trajectory-causal-attribution verified by the PulseGate indexer
Other apps tracked under the same category.
