The Problem
A busy Magento store generates a constant stream of entries in var/log and var/report. Most of it is repeat noise — the same warning logged hundreds of times — which trains a dev team to ignore the log directory entirely, right up until the one entry that mattered gets missed.
The Approach
The module is an open-source Magento 2 component that scans the log and report folders on a schedule or on demand via CLI, filters by severity, and uses a tracking file (notified.json) to deduplicate — so a team is alerted once per newly discovered issue, not once per occurrence.
Where AI Fits In
AI is optional and provider-agnostic: incident summaries can be generated with Google Gemini, Hugging Face, OpenRouter or Groq, with automatic fallback (Hugging Face falls back to its router API if the standard endpoint fails). Every AI request and response is logged separately for troubleshooting — the AI layer is auditable, not a black box bolted onto alerting.
The Outcome
The module ships as a standard Magento 2 component (app/code install, admin-configurable severity thresholds and recipients) and is open source on GitHub. It's a working example of AI used for what it's actually good at here — turning a raw error into a readable summary — while the deduplication and alerting logic stays deterministic and debuggable.