Technical
The Bare Metal Setup Experience: From Crate to Prompt in Under an Hour
A sealed appliance is only as good as its unboxing story. What actually has to happen between a locked case arriving on a loading dock and an authorized user typing a first prompt.
· 7 min read
Cloud AI's onboarding story is a login page. A sealed appliance's onboarding story is a crate on a loading dock, a receiving clerk with a clipboard, and, somewhere down the hall, a facility that has no persistent internet connection for the thing inside to phone home to. Those are not the same problem, and pretending they are is how sealed-hardware projects turn into six-week integration efforts instead of an afternoon.
The promise of an appliance model is that the hard problems get solved once, in the factory, instead of once per deployment, in the field. That promise is only real if the physical and procedural path from delivery to first authenticated use is actually short, and short in a way that holds up under the constraints a regulated or air-gapped environment imposes, not short in the sense of a demo that skips the parts that would be inconvenient on a real site.
What "under an hour" has to survive
The number is not a marketing figure so much as a design target, and it is worth being precise about what it has to survive to mean anything. It has to survive a receiving process that was not built with a sealed compute appliance in mind. Most loading docks are set up for pallets of consumables, not tamper-evident hardware that needs to be logged, inspected, and moved without breaking a seal that proves nothing happened to it in transit. It has to survive a network environment that, by the buyer's own requirement, may offer nothing to connect to: no DHCP handing out an address, no outbound path to a vendor's activation service, sometimes no network at all beyond what the appliance brings with it. And it has to survive the reality that the people doing the unboxing are frequently not the people who understand the cryptography underneath. A systems administrator or a facilities technician, not a security engineer, needs to be able to get from crate to working system without a support call.
Design for those constraints from the start and the setup path gets short because there is less left to do on site, not because steps were skipped. Design for a friendlier environment and bolt air-gap handling on afterward, and "under an hour" quietly becomes "under an hour, assuming the network cooperates and someone senior is in the room," which is not the same claim.
Tamper evidence comes before power comes on
The first step in any credible bare-metal setup is not technical, it is inspection: verifying that the seal the appliance shipped with is intact before the unit is ever powered on. This is the point where the physical security properties described elsewhere in the E31 lineup's design actually get exercised for the first time outside the factory. A tamper-evident enclosure is only worth what it is worth if someone on the receiving end actually checks it, and checks it against a known-good record of what the seal looked like when the unit left manufacturing, rather than trusting it on sight.
That verification step is deliberately simple to execute, a visual and mechanical check against a reference, not a specialized inspection requiring a security clearance to interpret, because the goal is a control a receiving clerk can actually perform correctly, every time, rather than a control that is theoretically strong but practically skipped because it is inconvenient. A tamper check that gets waved through because it was too cumbersome to do properly is not a control at all.
Bootstrapping identity without a network to lean on
Most enterprise hardware bootstraps its identity and trust relationships over a network: it calls home, it pulls a certificate, it registers itself with a fleet management service somewhere in the cloud. A sealed appliance destined for a genuinely air-gapped environment cannot assume any of that is available, and often the environment's own policy forbids it even if it were.
The alternative is provisioning identity and cryptographic trust material before the unit ever leaves the factory floor, so that the on-site steps are about establishing local operator access and local network configuration, the things that are specific to the deploying site and cannot be known in advance, rather than about negotiating trust with an external party the site cannot or will not reach. First boot on site should be closer to unlocking something that already knows what it is than to enrolling a blank machine into an identity system it has never talked to. That distinction is what keeps the process from silently depending on connectivity the deployment environment was never going to provide.
The role of the local operator
Somewhere in the sequence, a human with the right authority has to take custody of the appliance and establish that they, specifically, are the one allowed to configure and later use it. In a cloud product this is an email and a password. In a sealed appliance intended for defense, government, or regulated enterprise use, it has to be stronger than that by construction, because the appliance may never again check in with anyone who could revoke a credential remotely.
That generally means the initial administrative handoff is local and deliberate: physical possession of the unit, a credential or key material handled entirely within the facility's own control, and an enrollment step that does not depend on any service outside the room. It is slower, by design, than a cloud signup flow. But the hour budget accounts for that, because this is the one step in the process that is inherently manual and site-specific, not a candidate for automation away.
Configuration that starts narrow, not blank
An appliance that arrives as a genuinely blank slate, no defaults, no preloaded stack, every parameter to be decided fresh on site, is not actually faster to bring up, even though it sounds more flexible. Every open decision is a decision the deploying team has to make correctly, under time pressure, usually without the vendor's engineers in the room. The realistic path to a short setup window is the opposite: a small, sensible set of defaults established in the factory, with the on-site configuration surface limited to the handful of parameters that are genuinely specific to that deployment: network addressing, local operator enrollment, and whichever integration points the buyer's environment requires.
This is where Substrate's role matters underneath any of the individual products: the memory, retrieval, and integrity layers do not need to be assembled from scratch at every site, because they are the same underlying platform whether the appliance in the crate is running Forge's coding-assistant workload or Czar's training and fine-tuning workload. The setup experience differs at the edges, what gets connected, what gets loaded, who is enrolled, not in the foundation underneath it.
What "prompt" actually means at the end of that hour
The end state worth aiming for is not merely a powered-on machine. It is an authenticated, authorized user able to issue a first real request and get a real response, inside the environment's own network boundary, with no step in between that quietly depended on connectivity the site does not have. That is a meaningfully higher bar than "the hardware boots," and it is the bar that determines whether the appliance model actually delivers on its core promise: that the difficult work of building a secure, sovereign system happens once, in a controlled setting, so that the work remaining for the buyer is measured in minutes of configuration rather than weeks of integration.
Getting there inside an hour is not a single trick. It is the compounding effect of a tamper-evident enclosure that is easy to check correctly, an identity model that does not need a network to bootstrap, a local-operator handoff that is deliberately manual only where manual is actually necessary, and a configuration surface narrowed to what the deployment site genuinely has to decide. None of those pieces is exotic on its own. What is uncommon is designing all of them, from the outset, around the assumption that the loading dock they arrive at may have no path back to anywhere else at all.