Enterprise AI means paying per seat, per month, in dollars, forever. Kamry One is an appliance that runs large AI models in your own building: buy it once, the whole team uses it. Kamry OS makes it usable — running in five minutes without an ML engineer, with ready-made agents for your sector. Already own a capable machine? We license the OS for that instead.
The appliance.
01 / PRODUCT
Customize Your Appliance
{{ currentChipLabel }}Two memory architectures were costed. Only the co-packaged one is procurable today, so it is the only one priced. Choose a compute tier, then size NVMe storage for your corpus.
It needs a carrier we design around a discrete LPDDR5X array, a 14–16 layer HDI substrate and a signal-integrity programme at 8533 MT/s. We revisit it for Batch 02 when discrete LPDDR5X falls below about $6/GB. Until then we quote it as an engineering programme rather than sell a checkbox we cannot honour.
Anyone can buy the silicon.
Every other route to local AI ships a Linux box and a command line. Kamry ships the working office.
| DGX Spark$4,699 | Jetson Thor$5,499 | DIY buildPARTS COST | Tiiny AI$1,399 | Kamry One$9,450 | |
|---|---|---|---|---|---|
| Silicon that can hold a 100B model | |||||
| Models resident and answering on arrival | |||||
| Serves a whole office, with seats and roles | |||||
| Sector agents ready to install | |||||
| Audit log and residency attestation | |||||
| Support on the ground in Africa |
Five minutes is the operating system, not the silicon. Kamry OS is what turns any capable machine into the appliance — the same Builder, the same agents, the same audit log, whichever box it lands on. Bundled with every Kamry appliance; licensed separately for hardware you already have.
Target for a pre-flashed unit with resident models; corpus indexing continues in the background.
Bars are resident weights at 4-bit, except GLM-5.2, shown at its 2-bit dynamic quantisation. Anything larger still runs but pages its inactive experts from local NVMe — a trillion-parameter MoE such as Kimi K2.6 needs roughly 590 GB resident at 4-bit.
- {{ d.title }}
The Agent Store covers the common cases. Builder covers the rest: a visual canvas for wiring your own agent, tuning it on your documents, scoring it, and publishing it to the firm. It is also how we deliver custom agent builds — the same studio, operated by us.
Kamry is pre-prototype by design. The Builder and the OS are specified and being built — that is what this round funds, with first deliveries in Batch 01, 2027. We would rather say so than show you a screenshot of software that does not exist.
Model footprints are indicative, with context overhead, and will be replaced by measured figures from production units. Open weights remain the property of their publishers; Kamry ships the runtime, not the models.Yet.
Where it goes first.
04 / SECTORS{{ s.blurb }}
Questions.
05 / FAQ{{ f.a }}
Talk to the founders.
06 / CONTACTPiloting Kamry in your firm, clinic or agency? Write to us directly — we answer every email ourselves.
business@ajianlabs.com