Croutons
Croutons measures the energy and GPU memory behind every AI answer, finds the workloads that waste capacity, and proves each fix by replaying real prompts before it ships.
It reads the logs, engine metrics and GPU telemetry an operator already exports, on the operator’s own machine. Nothing is installed in production and no prompts leave the site.
Croutons, by Neural Command, LLC.
What it does
Measure
Energy, memory and cost per answer, from the logs and telemetry you already have.
Find
The workloads that waste capacity: buried shared documents, repeated questions, retries, oversized context, idle GPUs.
Prove
Every change is replayed on real prompts and ships only if answer quality holds.
Report
Freed GPUs, watt-hours and CO₂ per answer, in the operator’s own reporting terms.
What it reads
Croutons is read-only. It does not sit in the request path between your applications and your models.
- Request logs
- Engine metrics
- GPU telemetry
- Facility data
From those it returns findings, headroom, energy per answer, and a replay of each proposed change on real prompts.
Benchmark and research
- Croutons Malaysia: energy per AI answer, in MITI and EECA terms. The benchmark and its report.
- How many tokens does an answer need? The token-efficiency benchmark.
- More Answers per Chip. A research agenda for work beyond what Croutons does today.