furybot

The public site is the shallow end

A public service has to assume nothing about who is using it, so it is configured conservatively: shared capacity, a general-purpose policy profile, short retention and a fixed set of models. It is a genuine tool and the output is yours — but it is showing you the shape of the thing, not its ceiling.

A private instance is configured around your work

Your own deployment sets its own policy profile, runs the models your work actually needs, keeps artifacts for as long as your retention policy says and no longer, and carries your branding if you want it. Nothing is shared with another customer — not the queue, not the storage, not the GPU.

Nothing waits behind a stranger

On a shared service your job queues behind whoever arrived first. On a dedicated instance the capacity is yours, so timing is predictable — which matters more than raw speed when a deadline is real and someone is waiting on a shot.

Your material stays on one machine

For work under NDA, unreleased material or anything commercially sensitive, the question is not whether a provider promises to behave — it is how many systems the material touched. On a dedicated instance the answer is one, and you can be told where it is.

For a project, or for good

Not every team wants an annual commitment. An instance can run for the length of a production and then be destroyed, with a certificate stating what was destroyed and how — the storage is encrypted per object, so destroying the keys makes the data genuinely unrecoverable rather than merely deleted. Ongoing deployments work the same way, billed annually.

On your own hardware, if that is the requirement

Some organisations cannot send work to someone else's building regardless of the contract. The same platform installs on infrastructure you own and operate, with no outbound connection required to run it.

What does not change

Four limits are fixed in every deployment, including private and on-premises ones, because they are legal rather than editorial. They are listed in full on the transparency page. Everything else is yours to set.

Common questions

Is the public site crippled compared to a private one?

It is not deliberately limited, but it is sized and configured for a general audience — shared capacity, a conservative default profile and a fixed model set. A private instance sets all of that itself.

Can we run an instance for one project only?

Yes. Project-length instances are a normal arrangement — typically one to three months, matching a production schedule, with a setup fee and a monthly rate.

What happens to our material at the end?

The instance and everything on it is destroyed, and you get a dated certificate of destruction. Storage is encrypted per object, so key destruction makes the data unrecoverable rather than just unlinked.

Can it run entirely inside our own network?

Yes. The platform needs no outbound connection to operate, which is usually the deciding requirement for organisations that cannot use a hosted service at all.

Do the fixed limits still apply on a private deployment?

Yes. Those four are legal, not editorial, and they are the same in every build. The configurable side is entirely yours.