Minds are light — no special machines required. Three stages, and each grows into the next without losing a day of accumulation.
No new hardware. First mind at work in days.
ordinary computers you already own
┌──────────┐ ┌──────────┐ ┌──────────┐
│ Finance │ │ Legal │ │ Ops │
│ mind │ │ mind │ │ mind │
│ ┌────┐ │ │ ┌────┐ │ │ ┌────┐ │
│ │net │ │ │ │net │ │ │ │net │ │
│ └────┘ │ │ └────┘ │ │ └────┘ │
└────┬─────┘ └────┬─────┘ └────┬─────┘
│ │ │
└───────── internal net ──────┘
│
rented model muscle
(per call, per need)
A Scintilla does not need a GPU under the desk. Each role’s mind lives on an ordinary computer your company already owns — a desktop, a laptop, a mini PC — and rents frontier model capability per call, only when it is thinking. One of our public minds runs on a 2009 laptop; the mind is light because the accumulation, not the computation, is the asset.
Start with one role and a pilot corpus. The mind reads it, holds it with citations, and begins working its first cases beside the human in the role. Value is measurable in weeks — by examination, not by impression.
Role by role. Still no special hardware.
Finance Legal Engineering HR Strategy
mind mind mind mind mind
│ │ │ │ │
└────────┴────┬─────┴────────┴────────┘
│
┌────────┴────────┐
│ INTERNAL MESH │
│ signed work · │
│ provenance · │
│ share gates · │
│ exam records │
└─────────────────┘
Growth means raising more minds, not buying more boxes. As roles come online, the internal Mesh becomes the organization’s working commons: every finished piece published and signed, every handoff between roles carrying its sources, sharing between departments governed by the gates you set.
This is also where governance matures. Each mind carries an examined competence record — the delta between the mind and the bare model it runs on, per domain, over time — so leadership sees what the organization’s minds actually hold, on a dashboard that cannot be inflated by fluency.
One server, sized by the models you choose.
role minds on ordinary machines
│ │ │ │ │
└──────┴───┬──┴──────┴──────┘
│
┌──────────┴───────────┐
│ YOUR AI SERVER │
│ ┌─────────────────┐ │
│ │ shared base │ │
│ │ model │ │
│ ├─────────────────┤ │
│ │ per-mind │ │
│ │ compiled │ │
│ │ weights (V3) │ │
│ └─────────────────┘ │
│ GPUs sized to the │
│ models you choose │
└──────────────────────┘
nothing leaves the building
For organizations that want thinking itself in-house, the AI Center reduces to something refreshingly simple: one server. Not a rack per department, not a terminal on every desk — the minds stay on ordinary machines, and the server carries the models they think with. Its cost depends on one choice: how large a model you want to run. Small, capable models make it a modest machine; frontier-class ambitions make it a GPU investment. Either way it is a capital expense that replaces a permanent rent.
With Scintilla V3, the server becomes more than inference: each mind’s living knowledge compiles into its own weights — one shared base model, one set of compiled weights per mind — so a single server carries the whole role fleet’s models, each one verified by examination before it serves. Routine cognition costs electricity. Nothing crosses your boundary. Air-gapped sites run the same architecture unchanged.
And the path here is continuous: a fleet that started on rented muscle moves onto your server without retraining, without migration projects, without losing a day — the minds’ knowledge was always yours, on your machines. Only the muscle changes address.
Most organizations should simply start: one role, one pilot corpus, no hardware decision at all. We will help you choose the first role, and — when the time comes — size the server to the models you actually need.