The knowledge layer for enterprise AI.
Cornerstone turns what your organisation knows into cited facts your AI works from. Answers trace back to what your business actually wrote, and the same token budget covers more work as your knowledge grows.
On a public benchmark
Seven methods, the same documents, the same model. Cornerstone found the right document 96.7% of the time against the best alternative’s 85%, on 98% fewer tokens than reading them — and was the only one of the seven that could tell when the answer was not there at all, declining 41 of the 49 questions the documents could not answer.
The scaling wall
Enterprise AI works in the pilot and breaks at scale.
Point it at the real estate — twenty years of documents, four systems called billing, the decision nobody wrote down — and cost rises, accuracy falls, and nobody can tell you why an answer said what it said.
Token efficiency
The same budget covers more work
Flat retrieval sends more text through the model with every question, so the more your organisation knows, the more each answer costs. Cited facts break that link, and your token budget reaches further as the knowledge grows.
Productivity
Exactly the context the question needs
The AI receives the facts that answer the question, cited, instead of a pile of documents to sort through. Less to read, less to get wrong, less to redo.
Trust
Every answer traces back to something you wrote
Typed facts carry their source, their author and their date. When something turns out to be wrong, you can find every decision that rested on it.
Proof
Seven methods. Same documents. Same model.
Every method we could build, measured the same way on the same corpus. Bring your own awkward questions and we will run them with you.
Measured, not asserted
Seven methods. Same documents. Same model.
How much less you spend for each answer you can actually use. Query tokens only — the one-off cost of building the index is excluded, because on your own model it is not billed per token. Each bar is what that method costs you that Cornerstone does not.
- Cornerstone vs File searchReading whole documents99.7%
- Cornerstone vs Vector RAGSearching by meaning20.2%
- Cornerstone vs Hybrid + rerankKeywords + meaning, re-sorted18.2%
- Cornerstone vs Hybrid RAGKeywords + meaning · best alternative13.8%
Not on this measure: GraphRAG. A knowledge graph answers from its own graph rather than by handing back a source document, so “did it find the right document?” has no answer in these terms — and without that there is nothing to divide the spend by. It is on the other three measures, and on the trap corpus it captured 7 of 10 planted facts against Cornerstone’s 10.
13.8% cheaper per correct answer than the best alternative, 20.2% cheaper than searching by meaning, and 99.7% cheaper than reading whole documents.
How it wins
Per question these methods look level — Cornerstone is 1.5% cheaper than vector search and nobody should buy anything on that. The gap opens once you count only the questions that were actually answered. Vector search finds the right document 78.3% of the time, so a fifth of what it spends buys nothing; Cornerstone finds it 96.7% of the time. Same spend, more of it landing: 2,421 tokens per correct answer against 2,810 for the best alternative, 3,033 for vector search, and 861,225 for handing the AI whole documents to read. Measured directly rather than derived, the three-method report puts the vector gap wider still, at 26.0%.
Why it matters to you
This is the number that behaves like a bill. You are not buying questions, you are buying answers somebody can act on, and a method that is fractionally cheaper per question while missing one in five is charging you for the misses twice over — once in tokens, and again in the time it takes somebody to work out the answer was never there. One caveat stated plainly: building the index is a real one-off cost, and on a hosted model it is a real token bill. Run your own model, as Foundry customers do, and it is an overnight job on a GPU you already own — 3.8 hours for 2,201 documents — with no per-token charge at all. That is why it is excluded here, and it is the only thing excluded.
Knows when there is no answer — 41 of 49
Every other method scored no — and not narrowly. A search always has a closest match, even when nothing in your documents answers the question at all. So it hands back its best guesses, and the AI writes them up with exactly the same confidence it would use if the answer were really there. Nothing in what you read tells you which just happened.
The platform
One knowledge layer. Two places your people work.
Your organisation's knowledge becomes facts your AI can be held to. Your people meet those facts in Workbench and your leadership sees them in Habitat.
01
It learns what your business knows
Documents, decisions and conversations are distilled into individual facts — each one carrying its source, its author, its date and what it applies to.
02
Your team works in Workbench
Ask, draft and decide with that knowledge already underneath the conversation, on whichever AI model you have chosen. Nothing to wire up first.
03
Leadership sees it in Habitat
A live view of everything the organisation has running, built from the systems already in place — so it is current because it is live, not because somebody updated a deck.
EU client data stays in EU regions, and Frankfurt is the default. The residency addendum has to be attached to any EU contract — that rule was set by Legal in March and the default region was an architecture decision in June.
One thing to flag: the rule predates the Milan region, so it may be worth confirming with Legal rather than assuming it still means Frankfurt only.
Select any building — a venture is a campus, one building per component.
Everything else is engine — coordination for people and their AIs is built in, and you can bring your own frontier model rather than ours. See how the platform fits together.
What changes
What changes in the first month.
Four things move, in the units the person signing it off is measured on.
Hours
Nobody re-explains the business to a machine
The context your people rebuild by hand every session is already there, and stays there. That time comes back immediately, and it compounds.
Risk
The confidently wrong answer stops reaching the work
A rule scoped to one region cannot come back as though it applied everywhere, and when the answer is not in your knowledge it can say so rather than producing something plausible — where every other method we tested could not say so at all.
Audit
Every answer traces to who wrote it, and when
Evidence assembled from facts that already carry their source and date, rather than reconstructed by people ahead of a review.
Capacity
More work inside the same AI budget
Cost per question holds inside a narrow band instead of swinging with whatever document happened to match, so it can be forecast — and the same spend reaches further as your knowledge grows.
What it is worth depends on what your work costs when it goes wrong — see it by industry and by department.
Sovereignty
Your knowledge stays yours.
For a regulated European enterprise this is not a preference, it is the condition of being allowed to use any of this at all. Run it hosted, in your own cloud, or on your own hardware.
Bring your own model
No token markup, and no dependency on ours.
Encryption and classification always on
Classification is how the access system works. There is no edition without it.
Air-gappable
Run the distillation locally and nothing needs to leave the building.
Start with one team, free.
Small teams use Cornerstone free. When it matters enough to widen, the conversation about deployment, sovereignty and scale is one we are ready to have.