Agentycs-in-a-Box

Six litres, 400 watts, and the whole platform inside it.

Talk to us about this Compare all eight patterns

What it is #

Agentycs-in-a-Box is a micro server: a six-litre ultra-small form factor you carry rather than install, drawing 400W, with a full-fat 96GB NVIDIA GPU and full server capability behind it. It will run from a battery. It needs no room, no mounting and no facility power feed, which is the whole point of it.

What it runs is the same platform, not a cut-down edition. This is a proper Agentycs PlatformInstance: Atom's lakehouse and hybrid search, Anima serving models on that GPU, the app runtime, durable workflows and the audit ledger, reconciled by the same operator, from the same declarations, as any other Agentycs instance at any size. An application validated in a managed cloud is the application that runs in a field.

It also answers the only sovereignty question that cannot be argued about. Model weights sit in the box's own artefact store, identity and secrets are minted inside it, and the audit ledger records locally, so pulling the network out changes nothing about what it can do. Capability you can carry is capability nobody else can switch off.

Who it is for

The work happens where there is nothing to install into, and frequently no mains power.
The decision needs an answer at the point it is made, not a round trip to a region.
Whoever uses the capability has to be able to carry it in themselves.
You want the same platform in the field that your people build on at home.

It goes in a backpack, runs off a battery, and needs nothing at the other end of a network.

What it actually is #

The reason this pattern has a page of its own is physical. These are the facts that decide whether the capability can go where the work is.

Form factor: A six-litre ultra-small micro server. Backpack-sized, and carried in rather than installed.
Power: 400W, and it will run from a battery rather than a facility supply.
Accelerator: A full-fat 96GB NVIDIA GPU, with full server capability behind it rather than an embedded board.
What it runs: A proper Agentycs PlatformInstance, Atom and Anima included. Not a cut-down portable edition.
Where it sits in the range: Apex is the entire hardware range; this is its micro end. The larger configurations in that range are a separate Apex offer.

Everything past that list is quoted rather than published: the exact configuration you would be offered, the environment it has been tested in, and what it serves for your models at your concurrency. A headline figure for one build is the wrong figure for the next.

Talk to us about a unit

What you get #

The whole platform runs here. These are the parts this pattern changes the shape of.

Six litres and 400 watts #

An ultra-small form factor that goes in a backpack and runs from a battery, on 400W. Those two numbers are what decide whether AI capability can go where the work actually is.

A full-fat 96GB GPU #

Not an embedded accelerator with a lower ceiling. A 96GB NVIDIA GPU with full server capability behind it, so the models worth running have somewhere to sit and Anima can serve them properly.

A real PlatformInstance #

Atom, Anima, the app runtime, durable workflows and the audit ledger, declared and reconciled exactly as they are anywhere else. There is no portable edition with the interesting parts taken out.

Nothing upstream to lose #

Weights are seeded into the box's own artefact store and identity is minted inside it, so it serves from the moment it powers up and keeps serving with the network unplugged.

The micro end of Apex #

Apex is the whole hardware range and this is its smallest member. Where a deployment wants a full facility footprint instead, the same platform runs on the larger Apex configurations without changing anything above the metal.

The operating model #

Nobody should discover who owns a failure while it is happening. This split is explicit before anything is signed, and it is the part of a deployment decision that outlives the architecture.

We supply: The integrated unit, its bill of materials and provenance, and the artefacts needed to rebuild it from nothing.
You operate: Where it goes, what powers it, who is standing next to it, and what network, if any, it is permitted to see.
The operator handles: Everything above the metal: platform reconciliation, tenants, model deployment and rollout, local backups.
We agree together: Acceptance, spares holding, the support path and the end-of-life route for each unit.