Rootly + Graphify turns incident history into a queryable knowledge graph

What it does

A new Rootly-focused fork of Graphify stitches incident history into a persistent, queryable knowledge graph. Built on the Graphify project — itself inspired by Andrej Karpathy’s LLM Wiki idea — the importer pulls incidents, triggered alerts, teams and the service catalog from the Rootly API, exports a local corpus, and builds graph.html, graph.json and a human-readable GRAPH_REPORT.md for exploration. It has been reported that the fork points Graphify directly at Rootly so you can see service heatmaps, team escalation maps, alert-to-incident funnels, and cross-service failure correlation in one interactive view.

Why does that matter? Because LLMs can stop reinventing the wheel on every query. Instead of summarizing a heap of past incidents ad hoc, the tool builds a structured graph that grows richer over time. Run a “deep” mode with Claude Code or Codex and parallel subagents infer cross-incident themes and recurring root-cause patterns — allegedly surfacing the “aha” moments ops teams live for: that single on-call person covering four schedules, that noisy alert source nobody fixed, or that pair of services that keep going down together.

How it works and how to try it

Installation is pip-simple: pip install "graphifyy[rootly]" and then graphify install (Claude Code) or graphify install --platform codex (Codex). Set ROOTLY_API_KEY in a .env, run graphify rootly (or non-interactively with flags) and the tool fetches data for a chosen 7/30/90‑day window, builds the initial graph and writes outputs to graphify-rootly-data/graphify-out/. Open graph.html in a browser, filter by team, severity, time range, or drill into the GRAPH_REPORT.md for suggested questions and “god nodes.”

This feels very much of-the-moment: knowledge graphs plus LLM agents are the hot combo in observability and incident learning right now. Want to stop tripping over the same outages twice? This is one practical attempt to make past incidents do real future work — and to turn noise into something you can actually ask a question about. The repo and instructions live on GitHub for anyone who wants to try it.

Sources: github.com/rootly-ai-labs, Hacker News