Say hello to

Meet Own 1
0%
0

Compute offers

One deployment layer across 32 providers.
0%
0

Cities and regions

Reachable in 60 countries.
0%
0

Open models benchmarked

Seven modalities, 1,224 runs.
0%
0

Lower compute cost

Up to, for the same workload.
0%
$0

Per token

No meter anywhere in the stack.
0%
0

Left the device

In one hour of local work on Own 1.

AI is becoming the most personal technology we have built. It learns how you think, holds what you know and acts in your name. Today all of that lives inside somebody else's platform, and leaving means starting again.

OwnX separates your intelligence from the machinery that runs it. Your memory, workspace, files, context, agents, history and rules stay with you, no matter what. Models, clouds and devices are parts you can swap. Your data. Your AI.

We bring the cutting edge of AI infrastructure and AI platform-as-a-service into one stack: TPM2 hardware security, unlimited tokens, key-based private networking, reproducible NixOS apps, and routing across 32 providers and 482 regions. All of it open source.

The first era of AI was about models. The next will be about who owns the intelligence they create.

Scalable by design.

Start on one machine on your desk. Add your own servers. Reach 482 regions when a job needs them. The same system the whole way up.

Private by default.

Your data, memory and history are encrypted with keys you hold. Open models you pick do the work, and nothing you send trains anyone else's.

Independent by architecture.

Your intelligence is separate from the model, the cloud and the device underneath it. Swap any of them and keep everything you built.

Five layers of innovation.
One owner. You.

Nobody else builds all five. The computer on your desk, a platform for every open model, an operating system for agents, a cloud that routes across the world's compute, and the price index underneath it.

Own 1.
A mind of your own.

A powerful personal AI computer. Own your data. Own your AI. Two local models and eight agents run on a silent machine 158 mm square, with no subscription and no per-token cost.

Reserve Own 1Full specifications
Own 1 from the front right: a machined silver body with the OwnX mark on top
Own 1 front face under a glowing Own1 wordmark
Own 1 rear face with its ports under four vents
Four Own 1 units stacked into one column
Own 1 standing in its docking station beside a desktop graphics card

2 + 8Local models and agents at once
99TOPS of on-device AI
8 TBPrivate encrypted vault
75 WFlat out, an eighth of a gaming PC
$0Per token, no meter, no rate limit

OwnX

Open multimodal intelligence platform

One intelligent platform. Every open-weight model. Cloud or local. Run anywhere.

Illustrative AutoRoute
Layer 01Applications and agentsPick the kind of work.
Layer 02Open-weight modelsOwnX picks the model that fits.
Layer 03Infrastructure of your choiceOwnXCloud picks the machine.

OwnX qualifies the model. OwnXCloud qualifies the machine. You keep the project, the memory and the bill in one place.

Explore OwnX

Open-weight models, running on OwnXCloud.

Every model is benchmarked first, then placed on the cloud, region and hardware that runs it best for the price. Pick a model to see where it runs.

Models across cloudsIllustrative placement

Swipe sideways to see all 11 clouds

Where it runs bestLive index, Aug 2026

Cheapest region
Median price
Time to first tokenMeasured
0.9 smedian

Our slowest moment, 2.2 s, beats every rival's. 2,000 requests per provider, same prompt and region.

One model, one yearLive index
80%less

The same 8B model at the same token volume for twelve months.

Quality against the frontier baselineMeasured
5 of 7task classes

We claim five. We publish the two we lose.

Which model sits on which cloud is shown as an illustration of routing. Regional prices are indexed medians across providers with capacity in that region, August 2026. Quality is a blind pairwise eval scored by keyword heuristic, not a human rating.

Introducing unlimited tokens.

Rented AI bills you for every word it reads and writes. On OwnX the model runs on hardware you own or compute you choose, so the bill stops following the token count.

The meter, running since you opened this pageYour inputs
0

Tokens your team has used while you read

Metered API$0/mo

Rises with every token. Agents make it rise faster.

OwnX$0/token

Own 1 costs about $8 a month in electricity, whatever the count.

Rent it or own it, over 36 monthsModelled
$15,388

stays with you by month 36.

Rent $19,800 Own $4,412
Five subscriptions or onePublished prices
$85saved every month

Tap the tools you pay for today. OwnX does all five jobs for one price.

You pay now$110
OwnX$25
Hours of work before a limitMeasured
3 of 5have cut you off

Drag through a working day. Five paid assistants stop one by one. OwnX has no cap to reach.

One batch job, five decisionsMeasured
$10for the same job

Tap each step to apply one more decision.

Own 1

Buy the machine once. Two local models and eight agents run on your desk around the clock with no per-token cost and no rate limit.

See Own 1

OwnX

No machine yet? Open OwnX in a browser and take an 8-hour or 24-hour pass. Use as many tokens as the work needs while the pass is live.

See OwnX

Rent line: $110 per person per month, from published subscription prices plus overage and egress on our own invoices. Own line: one Own 1 shared by up to ten people, plus electricity and a replacement fund, hardware amortised at zero residual value. At one or two people renting stays cheaper inside 36 months, and the chart shows it. Limits measured on paid tiers, five runs each.

One intelligence.
Eight kinds of work.

Every request goes to the open model best suited to it, on your machine when it fits and on OwnXCloud when it does not. One workspace, one memory, one bill.

Model
Ran on
In practice

Outputs above are illustrations of each workspace, not recordings. Speeds quoted are best measured results from the Own 1 benchmark programme, 20 May to 27 June 2026.