Cited · Corroborated · Realtime · Private
Orrery reads every document and joins the threads between them. Ask anything, and the answer comes back fast, with the evidence and anything that contradicts it.
About us
Kinoto designs and operates AI systems for the organisations that can least afford to leak, or to be wrong: government bodies, healthcare providers, regulated industries, and anyone who has to answer for what their AI tells them.
We work on your infrastructure or run our own; either way, we take your raw data, shape it into something a model can actually use, train small focused models for the questions that matter to you, and operate the whole pipeline so every step stays auditable and under your control.
In our line of work, a confident answer you can't check is worse than no answer at all. So we built AI that shows the source behind every claim, surfaces where your own documents disagree, and lets you prove later exactly how it got there, on data that never leaves your control.
Values
Every answer traces back to its source, and every step the system takes is recorded by default. So when it's questioned, you can show exactly what happened, instead of taking our word for it.
Your data belongs to you. We only ever borrow it, with access that's bounded, recorded, and never wider than you've allowed.
Walled off in a space of its own, never pooled with another customer's, and never used to train a model or shown to anyone else.
Built for environments that can't afford to fail: tight permissions, signed releases, and the kind of steady, predictable behaviour auditors expect to see.
Our products
Available
Point Orrery at your documents: a book, a contract, a transcript, a stack of meeting notes. It reads them all, answers your questions in plain language, quotes the passage behind each answer, and flags where those documents disagree.
See OrreryOn the roadmap
Helios sits between your raw sources and the systems that need clean data. It gathers everything in (PDFs, recordings, scans, spreadsheets, live feeds), tags and organises it as it lands, and lifts out the facts that matter. Anything it isn't sure about goes to a person to confirm.
Express interestOn the roadmap
Run your models, pipelines, and training jobs on dedicated machines built for sensitive work. We keep the hardware running; the data stays yours alone.
Express interestUnder the hood
Eight engineering choices sit behind every Kinoto product, two for each value above: provability, privacy, containment, and security. None are flashy on their own; together they're why those guarantees actually hold.
01
Provability
Switch it on and every output gets signed, naming the exact code and model version that made it. Months later, that signature still settles which version said what.
02
Provability
Answers, citations, admin actions: all of it is written down as it happens. When an auditor asks why the system did something two years ago, you replay the actual run instead of reconstructing it from memory.
03
Privacy
Training has to read your text in the clear, so the only eyes on it are ones you've approved. By default that's your own walled-off space; for the touchiest work, a hardware enclave even our own engineers can't see into.
04
Privacy
Your terminology and house style get baked into a model that's yours and no one else's. It never trains the next customer's model, and the weights don't leave the building without your say-so.
05
Containment
One model per job, not one giant model for everything. Each is small enough to answer in milliseconds, cheap to run, and simple enough that you can say exactly what it does and doesn't touch.
06
Containment
Dashboards track drift, latency, and load as it happens. What leaves the system is counts and averages, never the contents, so keeping an eye on the work never opens a side door out of it.
07
Security
A model that can reach a database or an outside service can also be talked into misusing one. So every call runs through a checkpoint: logged, and limited to the short list you've allowed.
08
Security
Permissions are fine-grained and revocation is instant. Hit stop and the whole line halts, not just one process while the work quietly carries on next door.
Use cases
These are the patterns Kinoto is built around: scenarios where the data is too sensitive, too regulated, or too valuable to hand to a general-purpose AI.
Manufacturing · Legal · Pharma
Your contracts, specifications, and regulatory text, opened up to your team, your clients, or both, without any of it leaving your own private space on our servers. Every answer points back to the clause it came from, and nothing is ever sent to a third-party AI.
Audit · Banking · Healthcare compliance
Thousands of filings and case files, opened up to whoever's authorised to read them (staff, auditors, regulators, customers), with every answer tied back to its source filing, contradictions between them flagged before they become findings, every query logged, and every decision replayable years later when the audit comes round.