So What. Labs
Coming soon
Lore.
If AI was done right.
Enterprise document intelligence. Lore takes the sprawl your business actually runs on — contracts, transcripts, policies, decks, shared drives, ticket threads — condenses every document to what it really says, and organizes the whole estate under one ontology. Then it answers questions about it, with the source attached.
Point it at
the mess.
No cleanup project first.
Enterprise knowledge doesn't sit in a tidy repository. It sits in twelve-year-old shared drives, a wiki two people maintain, mail threads, call transcripts, and four versions of the same contract with FINAL in three of the filenames. Lore connects to where that already lives and starts reading — no migration, no renaming, no taxonomy workshop.
- Documents, spreadsheets, decks, transcripts and threads
- Connects to the systems you already run
- Duplicates and near-duplicates recognised, not multiplied
Ninety pages,
one clear record.
Signal kept. Padding gone.
Most enterprise documents are mostly boilerplate. Lore reduces each one to what it actually commits you to — the terms, the dates, the obligations, the decisions — and keeps a line back to the page it came from. The long version never disappears; it just stops being the only way to find out what's in it.
- Every condensed record stays traceable to its source
- Near-identical versions collapse into one, differences flagged
- Nothing is deleted — the original stays exactly where it was
Structure,
not a pile.
The part everyone else skips.
Search gives you documents. An ontology gives you a business. Lore maps what it reads onto the things your organisation actually deals in — clients, contracts, obligations, renewal dates, owners, systems, risks — and the relationships between them. Ask about a client and you get their commitments and their exposure, not ten files with their name in them.
- Entities and relationships modelled on your business, not generic tags
- The same structure across every source system
- Grows as the estate grows — no re-tagging exercise
Answers with
the receipts.
Checkable, or it doesn't count.
Ask in plain language and get a direct answer — with every claim linked to the document and page it came from. That's the difference between a tool your legal team tolerates and one they'll actually rely on: nobody has to take the model's word for it.
- Citations down to the document and the page
- Says so when the estate doesn't contain the answer
- Answers respect the asker's permissions, always
Built for the
security review.
The meeting most AI tools die in.
Lore runs as a hosted, isolated tenant with permission carried on the record itself — not bolted on at the interface. Access is filtered inside the query, so material somebody isn't cleared for is never a candidate for their answer. Every question and every source served is logged.
- Isolated tenant · encrypted in transit and at rest
- SSO and your existing groups, not a second directory
- Full audit trail: who asked what, and what it was shown
What your team
actually touches.
One hosted product. The ontology, the condensing and the retrieval do their work out of sight — these are the surfaces people use.
Ontology explorer
Walk the map of clients, contracts, obligations and owners. Open an entity and see everything the estate knows about it, gathered from every system at once.
Ask, with citations
Plain-language questions across everything the organisation has ever filed. Every answer arrives with its sources, so the claim can be checked instead of trusted.
Document briefs
Every file reduced to the terms, dates and obligations that matter — with a link back to the page. Read the brief in a minute, or the original when you need it.
Dates & obligations
Renewals, notice periods and commitments surfaced as a live list instead of something a person was supposed to remember. Nothing lapses because a file was buried.
Connectors & API
Reads from the systems you already pay for, and exposes the same ontology to your own tools — so Lore strengthens the stack instead of becoming another island in it.
Permissions & audit
SSO, your existing groups, per-record scope, and a log of every question asked and every source served. The security review is the first conversation, not the last.
Same method
we sell you.
Lore went through the identical four gates every client engagement does. If the process is good enough to charge for, it's good enough to build our own product with.
Kill the obvious version
The first build was a search box over a document store, and we scrapped it. Search was never the problem — nobody struggles to find a contract, they struggle to know what's in nine hundred of them. That finding turned Lore from a retrieval tool into an ontology product.
Test on the ugly corpus
We built against genuinely messy material — inconsistent naming, four near-identical versions, scanned pages, half-finished wikis. Clean demo datasets hide exactly the failures that matter once a real organisation plugs in its drive.
Score it, don't vibe it
Recall and citation accuracy are measured against a held-out set built to be adversarial — near-duplicate documents designed to trip retrieval into confident, wrong answers. Nothing ships if the number moves the wrong way.
Survive the security review
Then the unglamorous half: tenant isolation, SSO, permission on the record, an audit trail, and answers that admit when the estate doesn't contain them. Shipping is the feature.
If AI was
done right.
Not a demo that impresses a boardroom and dies in procurement. Structure before generation, citations before confidence, permissions before launch. That's the whole bet.
Bring us your
worst drive.
Lore is in active development and going to a small group of design partners first — organisations with a genuine document problem, who'll tell us the truth about what breaks. Tell us what you'd point it at and we'll put you on the list.
✉ allison.lamp@sowhat.company