ReplayLoom

ReplayLoom is a web app (with lightweight CLI agent) that records “debug snapshots” of ETL/pipeline runs so engineers can deterministically replay failures without re-pulling live sources. It captures inputs, schema versions, runtime config, environment metadata, and key intermediate artifacts (sampled or full, configurable), then provides a one-click replay in an isolated runner. The goal is not to replace your orchestrator (Airflow, Dagster, Prefect, dbt) but to sit beside it as a failure forensics layer. It generates a minimal reproducible package for a failed task, highlights diffs vs last successful run, and produces a shareable incident bundle for teammates. It also supports redaction rules and tokenization so sensitive fields can be safely stored. This is an AI + traditional app: AI helps summarize root-cause hypotheses and suggests targeted checks, but the core value is deterministic replay and artifact capture.