The queryable company
Every company is about to have more agents than employees. Almost none of them know how the company works.
Knowledge used to be a human problem
For twenty years the wiki was where knowledge went to die. People wrote pages once, nobody updated them, and the real answers lived in Slack threads and in the heads of the three people who had been there longest. It was annoying, but it worked — because humans could always ask a human.
Agents can't ask a human
Now the support agent answers 400 tickets a day. The finance agent closes the books. The ops agent writes the incident timeline. They are only as good as what they can query — and today, what they can query is a stale export of a wiki that was built for humans to write and nobody to read.
So every team builds the same brittle pipeline: sync docs into a vector database, pray it's fresh, hope the model doesn't make something up. No citations. No write-back. No idea who changed what.
What a queryable company looks like
- One source of truth, two audiences. Humans read it in a fast, clean app. Agents query it over MCP. Same data, same permissions.
- Every answer has a receipt. If a claim can't be traced to a page, a decision or a record, it isn't company truth.
- The why is first-class. Decisions get logged with their context, so “why is it like this?” has an answer.
- Agents write back. The things agents learn — incident timelines, updated metrics, new FAQs — flow back into the source of truth, attributed and reviewable.
- Freshness is visible. Every page has an owner and a verified date. Stale knowledge gets flagged before anyone quotes it.
Why now
Better models don't fix this. They make it more urgent: the bottleneck moves from “can the model reason?” to “does the model have the company's ground truth?” MCP gave every agent a standard way to plug into tools. The missing tool is the company itself.
That's what we're building. Join the waitlist, or start the free beta.
— The Groundtruth team