# What Nerve saved us, measured

> Last 30 days on Dubhlán's own Nerve instance, as published October 24, 2026. Spend is token usage priced at published API rates, not an invoice.

## The 30-day picture

- Spend at published API rates: $6,417
- The same usage without cached context: $26,650
- Measured saving: $20,233 (75.9% below that figure)
- Estimated saving, shown separately and never added to the measured one: $2,189
- Share of input tokens read from cache: 99.8% (90-day average 94.5%)

## By source

- Cached context (measured): $20,233. Every call records how many input tokens the model provider read from its cache. Cached tokens bill at about a tenth of the normal input price; the saving is the difference, priced per model at published rates.
- Work done without a model call (estimated): $2,172. Many agent steps (reading a file, querying the board, running a check) are handled by deterministic tools instead of a model round. We count those calls and value each at an assumed 1,500 input and 300 output tokens, priced at Claude Sonnet rates. The count is recorded; the per-call value is our assumption.
- Trimmed tool history (estimated): $16. Older tool results are summarised instead of re-sent on every round of a long task. We record the characters removed and convert at about 4 characters per token.
- Local models (estimated): $0. Tokens served by models you host yourself, valued at cloud prices. Our own instance runs no local models, so this is $0 for us.
- Smaller working memory (not measured): —. Not measured yet. Nerve does page context in and out, but we do not yet record a before-and-after we would stand behind, so we show a dash rather than a number.

## What carries over to your agents

- Always on: each agent's standing instructions and tool definitions are sent as an unchanging prefix the provider can cache, and routine steps run as deterministic tools.
- Optional, off by default: caching the running conversation as well. We switched it on for our instance on October 5, 2026.
- Your figures depend on your model mix, how long your agents work per task, and whether you already cache. The two-week assessment measures them on your own agents.

Book an assessment: https://dubhlan.com/waitlist?ref=savings