Uncapped capacity you'd otherwise meter
—
Per year of AI output the appliance can produce at ~22h/day — valued at what the same volume would cost on a metered API. For roughly the same budget you stop renting capped access and run it yourself, sovereign on your premises.
Annual SaaS spend
—
Capped seats, today
Appliance all-in / yr
—
Lease + subscription + power
Annual saving
—
SaaS spend − appliance lease, subscription & power
AI output capacity
—
≈ pages of generated output / day (~750 tok/page)
Local cost / million tokens
—
All-in lease vs metered API at right
Capacity at equal spend
—
Appliance tokens ÷ same-budget metered tokens
| Over 3 years | Per-seat SaaS | Omniscient appliance |
|---|
| Total cost | — | — |
| Usage ceiling | Capped — rate & token limits; idle outside sessions | None — unlimited, runs ~22h/day autonomously |
| 3-yr saving | — |
—
This is the quantifiable jump: cost holds roughly flat, but the work the AI can do is no longer limited by a plan — it's limited only by the hardware, which runs almost around the clock.
R&D Tax Incentive — — estimated refundable offset / year
—
What this model does & doesn't claim
- Capacity, not guaranteed productivity. It quantifies available AI work at equal spend. Realised productivity depends on how much of that capacity your teams actually use — which is exactly what removing caps unlocks.
- Local tokens ≠ frontier tokens 1:1. The bulk capacity runs on local open-weight models; for the hardest ~5–10% of tasks the box bursts to a frontier model (consented). The metered-API value is an order-of-magnitude comparison, not a quality equivalence.
- Throughput is your number. Tokens/sec depends on the model, GPU count and batching — benchmark the actual appliance; the default is deliberately conservative.
- Lease & subscription are modeled estimates (Business tier · 3-yr lease) — final per quote; every figure above is editable.