Where "why we did this" goes to die

Every team has a moment like this. Something breaks at 2am, and the person on call finds a config that makes no sense. Somebody, at some point, set a timeout to 800ms instead of the default 3000ms. Nobody remembers why. The person who set it left, or just doesn't remember, or is asleep.

So the on-call engineer does the thing everyone does: they either revert it (and maybe reintroduce whatever bug it was fixing) or they leave it alone and route around it (and the mystery compounds for the next person). Either way, the "why" is gone. It existed for maybe a week after the decision was made, living in a Slack thread or someone's memory, and then it evaporated.

Onboarding has the same shape. A new engineer opens the codebase and asks the obvious questions: why is auth done this way, why do we have two queues instead of one, why does this service call that one directly instead of through the gateway. The answers exist. They're just not attached to anything. They're folklore, passed down by whoever happens to still be around, in whatever fidelity survives a game of telephone.

This is what actually breaks during incidents and onboarding: not the docs (most teams have runbooks, ADRs, postmortems), but the connections between them. A postmortem references a decision doc. That decision doc references an earlier incident. Nobody wires that up explicitly, and no wiki search finds it, because search finds documents that match your words, not documents that explain your situation.

Here's what that looks like without context attached to your docs:

And here's the same situation once your docs are connected to an AI that can walk the relationships between them, not just keyword-match the text:

The difference isn't that the AI is smarter. It's that the underlying knowledge is structured as a graph of decisions, not a pile of markdown files that happen to sit in the same repo. A plain doc search MCP gives you the top loop, dressed up with an AI in the middle. It still can't tell you why, because "why" lives across documents, and file search only ever looks inside one at a time.

Rootr keeps your postmortems, ADRs, and runbooks as a workspace where those relationships are explicit, and it exposes that over MCP so Claude or ChatGPT can query the graph directly, not grep a folder. Ask "have we seen this before" and it can actually trace the answer back through prior incidents instead of hoping a keyword lines up.

If your team has ever spent an hour re-discovering something you already knew once, this is built for exactly that hour.

rootr.io