Positioning
Why it runs on your hardware
Not nostalgia for on-premise. Three specific consequences that a cloud platform structurally cannot offer you.
Your clients' data never becomes our problem
A cloud MSP platform holds every one of your clients' asset inventories, credentials, ticket contents and network topology. That is a concentration of risk, and your clients are downstream of a breach you cannot audit.
We receive ten fields and a signature. Not a policy — a contract enforced in code on both sides, published in full, and readable by you before anything is sent.
Your AI is not metered
Every cloud platform charging for AI has an incentive to charge you more when you use it more. That incentive shapes the product: features get gated, context gets trimmed, and the interesting question costs extra.
Genie runs on your GPU. There is nothing for us to meter, so there is nothing to gate. Ask him a hundred questions about one incident and it costs the same as asking none.
Our uptime is not your uptime
If we go offline, your monitoring keeps running, your alerts keep firing, your technicians keep working and your clients never know.
If we ceased trading tomorrow, you would lose updates and support. You would not lose the product, because it does not phone home to work.
And the ones people ask about second
- Air-gapped sites work fully. Licence, updates and usage move by file. Some clients require this and no cloud platform can serve them.
- Data residency is wherever your server is. No transfer agreements, no sub-processor list, no jurisdiction question.
- Nothing leaves your network. A question and its answer never cross the internet.
- Your retention policy is yours. Nobody deletes your logs at the end of a plan tier.
The honest cost
You provide the hardware, and one machine with a GPU. The spec is published and the card is $700–1,000 USD used, not a five-figure datacentre part.
You also own the backup of your own platform. We give you the tooling and the rehearsal schedule — but nobody else is going to restore it for you.
Published sizing, named cards, real prices.
On your hardware, in thirty days
Read-only, no card, and nothing leaves your building.
On speed, be realistic
The model runs on your hardware, so a question and its answer never cross the internet. That matters for what your clients' data touches. It is not why it feels fast.
A local 14B model on a consumer card answers in a second or two — comparable to a hosted model, and not the reason to choose this. A hosted frontier model is bigger and runs on enormous hardware; the round trip is noise against both. Choose this because nothing leaves your building, not because it is faster.
A fourth thing: an outage you can act on
When a cloud platform goes down, your whole practice goes down and there is nothing you can do but wait and tell your clients you are waiting.
A single node in your own rack is yours to restart — and while it is down, your agents keep running and keep queueing, and nothing is lost. If that is still not enough, it runs across three nodes with automatic failover.
Try it on your own estate
Thirty days, read-only, on your own hardware. No card, no call, and nothing to uninstall if you walk away.