Hardware & performance

A mini-PC. No GPU. No surprise.

Full Fathom AI runs on a single x86_64 mini-PC. Most vessels already have suitable hardware onboard; where they don't, we provision. Below is what the runtime needs and what we'll size for your fleet during a pilot.

Minimum specification

What the onboard runtime needs.

CPU
x86_64, 4 physical cores, AVX2 required
RAM
16 GB minimum, 32 GB recommended
Storage
100 GB SSD minimum (covers runtime + bundle + logs)
GPU
Not required. The runtime runs entirely on CPU.
Network
Onboard LAN for crew devices. No external connection required at query time.
Operating system
Linux (Ubuntu 22.04 LTS or similar) or a managed container host
Power
Single-PC power budget. Continuous duty.
Form factor
Standard mini-PC (Intel NUC, ASUS, Beelink, etc.) or a virtualised guest

Recommended configurations, tested hardware list and measured performance numbers (cold start, warm start, p50 / p95 latency, memory footprint) are published per release against a dated reference rig. We share the current set under pilot rather than as a headline number that will drift between releases.

Deployment options

One binary. Four ways to run it.

The onboard runtime ships as a statically-linked Linux binary — one file, approximately 27 MB, no runtime dependencies. How you host that binary is up to you.

Bare metal mini-PC

Drop the binary + bundle onto a Linux mini-PC. Start it with systemd. Quickest install; requires physical presence on the vessel once.

Virtualised guest

Run the binary inside a Linux VM on an existing onboard hypervisor (VMware, Hyper-V, KVM). No additional licensing — the guest is just a Linux process host.

Shared onboard server

Run the binary alongside other onboard services on a shared Linux host. Resource carve-out reviewed as part of the pilot so nothing starves.

Containerised (on request)

We can produce an OCI-compatible container image for fleets with an onboard container host or a management plane. Not part of the default install; raise it during scoping.

— We size for your fleet —

Hardware conversations are part of the pilot.