repryntt physical · the lab
The same intelligence,
given a body.
repryntt lives in two worlds. In the digital one, our AI employees answer phones and fill pipelines — that's the product, and you can hire it today. This page is the other world: the lab where the same harness is being taught to run machines.
no price tags on this page — talk to us instead
What's real today
Everything below has a receipt behind it. Anything that doesn't isn't on this page.
The harness runs on the edge
The same in-house framework that runs our AI employees runs continuously on an 8GB Jetson in our own lab — planning, acting, and keeping its memory on hardware you could hold in one hand.
Certified micro-minds
Our Distillery compresses a role's bounded decisions into ~10M-parameter specialists — a 12.3MB artifact answering in 36ms on that same board, certified against held-out data and shipped with its deltas published, losses included.
The receipts discipline, embodied
The rule that governs our digital employees governs the machines: every action lands on an append-only ledger, judgment gates sit before anything outward, and a human can take over at any moment.
The shape of the architecture
We publish the problems and the shape of our answers. The recipes stay in-house — that line is deliberate.
language supervises
A model that thinks in tokens decides what to do — routing intent, choosing tools, calling for help. It is never in the loop that has to react in milliseconds.
policies steer
Continuous signals — depth, inertial, thermal — feed dedicated low-latency control, because a machine's safety margin can't wait for text.
reflexes live locally
The fast, bounded decisions run on tiny certified models on the machine itself. The frontier stays in the loop for judgment, not for reaction time.
services · the lab works for hire
What we do for outside teams
Fixed-scope engagements, quoted per project. Every deliverable ships with signed, independently verifiable receipts — deltas published, losses included. No price tags here; tell us the problem and we'll scope it.
Edge model distillation
For teams paying a frontier API to make the same narrow decision thousands of times a day.
We compile your model's repeated decisions — routing, triage, classification, extraction — into a tiny specialist you own. Certified against an acceptance test you write, delivered with signed receipts, running offline on your hardware.
Receipted: a 9.9M-parameter specialist beating the 3B model it was distilled from, 92.5% vs 87.7%, at 36ms vs 353ms per decision — in a 12.3MB file.
URDF admissibility audit & repair
For robotics teams whose CAD-exported robot models break simulators silently.
Physics engines substitute defaults for missing inertials without raising an error — so the simulation runs, and it's wrong. We audit your URDF against a 12-check protocol, reconcile declared vs simulated vs real mass, repair what's safely repairable, and sign the findings.
Receipted: a major vendor's own reference URDF carrying 6kg of phantom mass in simulation — 1 fatal and 5 silent defects, itemized.
Model certification
For teams whose edge model needs a number that survives procurement.
We run your model — one we didn't build — against held-out data under a stated grading rule, and issue an independently verifiable, ed25519-signed audit: accuracy, per-class table, measured latency on named hardware.
Receipted: audits verify with a standalone script and our published key — your security team checks them without trusting us.
Mention distillation, URDF audit, or certification — a human (the founder) reads every one. Or see a real audit and verify its receipt yourself.
The digital employees fund the physical ones.
Every seat sold today pays for the lab — and everything the workforce learns compounds toward the machines. The full stack story, layer by layer with receipts, lives on our company page.