What your own machines could take off the meter.
Send to my bossEstimate, not a quote
Off the meter$86,400 / yrLocal · your hardware
Off the meter$7,200 / moNo per-token charge on what your fleet serves.
Still metered$4,800 / moCloud · metered
The numbers you gave us
- Company-owned machines
- 300
- Monthly token spend today
- $12,000
- Non-human usage - agents, evals, embeddings, nightly jobs
- 60%
The arithmetic behind the estimate
Non-human work is what the fleet is eligible to serve. It carries what it can; the overflow, and every human-facing request, stays with the cloud providers.
non-human work = $12,000 × 60% = $7,200 / mofleet capacity = 300 machines × $40 = $12,000 / mooff the meter = lesser of the two = $7,200 / mostill metered = $12,000 - $7,200 = $4,800 / mo
The workload is the limit here, not the fleet. 300 machines carry all of the non-human work, with capacity to spare.
the number kibbu has not measured
$40/machine/month is a working figure, not a benchmark. Everything else on this page is your own input or arithmetic on it. The benchmark campaign replaces this number with a measured one, and a pilot replaces it with yours.
What a pilot would settle
- Whether $40/machine/month is anywhere near right for your hardware.
- Per-request attribution: what ran locally, what fell back to cloud, and what the fallback cost.
- Which models your machines can serve, and how the fleet behaves under real load.
- Whether anyone whose machine is enrolled notices. The target is that they do not.
We validate every number with you during a pilot.
Where to go from here
The call button opens a short email to hello@kibbu.io with your estimate link in it.