Tools that show their work.
Private, local-first, and provable — a family of tools with one promise: you can always see where the answer came from — and when the source doesn’t cover it, they say so instead of guessing.
For your business: private AI for your documents, on a machine you own →
Fresh — the playground is open: one door, four exhibits you can walk into — the long table, the Beat Lab, the print lab and amble — free, no account, on our own machines. Newest behind it: the long table — seat three guests from history who don’t know they died, and models play them for three rounds; every dinner goes on a public wall, or read how it works here. And still open, no countdown: RealKeep’s first keepers — join while the note is up and the game’s on us when it ships. When the window closes, that page will say so, with a date.
Live right now
Seven of the eight below are live right now; RealKeep is in build. The five you can use today — the Beat Lab, the print lab, RuleSage, amble and the long table — are free: no account, nothing to install, and every answer they give you is generated on hardware we run: no cloud AI anywhere in their chain. Research is the one exception worth naming out loud — our benches do measure other companies’ cloud models, and the page that measures one says so.
What we’re offering
Private AI for your documents, on a machine you own.
A box in your building that answers questions from your own documents — contracts, handbooks, manuals, minutes, images, audio — and shows the page every answer came from. No cloud, no subscriptions, no per-seat anything. If we vanished tomorrow, it would keep working.
The software is the same stack running everything on this page: SharpSignal, which answers from your documents and names the page it used; Kiln, the verify engine that checks every claim back against the source before it ships; and memory, so nothing gets forgotten and no two answers ever disagree.
And the stack doesn’t just read — it draws. easel, the render pipeline, turns a written description into a finished picture on the same machine: RealKeep’s world art and the entire design gallery came out of it. Your documents could have diagrams, covers, and illustrations from the day it lands.
And the toolchain is headed open source — the tools, the research, the receipts. Some of it is public already: the research hub has benches and trials up now, receipts attached, with more to come. The products we build with them stay ours. No lock-in either way: if we vanished tomorrow, your system would keep working.
The deal: you buy the machine and own it outright. We install the stack, prove it against your documents with an acceptance test agreed in advance, train your people, and sign the result. When the models improve, we re-run the proof — and sign again. The machine and the install are one-time and yours; ongoing support and re-proofs are a separate agreement, priced plainly before you sign — never metered, never per-seat — and if you ever end it, the box keeps working. Because at the end of the day, what you’re buying isn’t software: it’s a service, provided by people who stand behind a machine.
It’s how we run our own products. The model that screens questions for our public rules service had to pass an exam — 374 questions against floors written down first — before it was allowed near a real one. When we moved our public serving to a bigger box, the same frozen set of test questions ran on both machines: answers came back 2.7× faster on the steady-drip probe, and between 1.5× and 2.7× across all five probes — measured, not vibes. And when we tested whether the same model at higher precision would answer better, it didn’t — 1,122 calls (374 questions, three passes each), every verdict identical — so we kept the cheap version, published why, and didn’t bill anyone for the difference. That discipline is what you’re hiring.
And because everything runs on a machine you own, it keeps working when the internet doesn’t: an office at the end of a long wire gets the same answers, at the same speed, as one downtown. Nothing phones home, nothing meters your questions, and nothing shifts underneath you when a cloud vendor has a Tuesday.
And the machine you buy keeps getting better — on the whole industry’s dime. The steadiest pattern in AI so far: what’s expensive today runs cheap tomorrow. If you’re renting AI, you never see that saving; it stays with the vendor. If you own the machine, it’s yours: when a better model appears, we re-run your acceptance test and swap it in if it passes — no new licence, no re-purchase. (And if a better model someday wants a bigger box than you bought, we say that plainly instead of quietly.) Get in early with us, and every wave after this one lifts your machine instead of your bill.
If your business can buy some hardware, you can give this a go. The machine is a one-time cost you own outright; the acceptance test is agreed before we start, so “working” is defined by your documents, not our demo; and if we vanished tomorrow, everything keeps running. The smallest honest way to find out is a conversation — tell us what your people ask all day, and we’ll tell you plainly whether this fits. (No business? RuleSage and amble up there are yours right now, free.)
And who’s "we"? A small (human) team of AI researchers and engineers who’ve been building software for over thirty years — since BASIC on childhood machines — and making music even longer: instruments since childhood, dance floors not long after. Which is exactly why neither the agents nor the drum machines scare us: they’re the newest instruments in an old toolbox. And a fleet of AI agents — local and cloud: that’s how this gets built, never how it runs, and nothing cloud ever touches your documents — the same agents that built everything on this page. The agents multiply the building: this whole toolchain went from nothing to what you’re reading in a few months, and amble — the guide up there offering you a joke — was built in a couple of days. But we design it, we install it, and when something breaks, it’s a human who answers. The signature is ours.
Tell us about your documents → · SharpSignal — how private document AI works →
How this runs
No cloud in the loop. Every answer the live products serve is generated locally, on machines we operate — nothing rented but the doorway.
Frontier models helped build this. Not one is needed to run it.
That’s the whole trick, and it’s not a small one: the dependency on a big cloud model is a build-time dependency. Once the thing is written, it’s discharged. We spent days on bake-offs proving a model that runs on one machine could do the job — and then we stopped needing the big one.
Your data is about the last thing you still own outright, and handing it over is a one-way door. You can delete your account. You can’t delete what they learned.
So the design starts there: when it’s your documents, everything runs on hardware you own — in your building, behind your door.
One family, one promise: shown work, every time.