An Incident Agent That Won't Repeat Old Fixes: Differential Reasoning Over Organizational Memory

Usual-Relief62 · reddit · 2026-09-29

Problem

The chronic ops bug: an engineer fixes an incident at 2am, writes a postmortem, and two months later another engineer wastes 45 minutes on the same fix. The author built OpsMind, an incident response agent backed by the memory layer Hindsight, to break that cycle.

System flow

Stack: FastAPI + React/Tailwind + PostgreSQL + Hindsight Cloud.

Core insight: differential reasoning over memory

Most DevOps RAG systems just find the nearest historical incident and regurgitate the old fix. OpsMind checks whether the historical fix is already applied: if a pool resize from 20 to 50 solved a past 500-error incident but dbpoolsize is already 50 today, a naive agent still suggests resizing; OpsMind detects this and pivots to slow unindexed queries, deployment regressions, or route-level connection leaks. The author shares the performdifferentialreasoning code that separates reciting memory from reasoning over it.

Related event: Hindsight as Agent Memory: Incident-Response Practices Surge, Benchmarks Mixed(21 posts)→

Original post →

More from coding & agent

coding & agent channel →