What “tokens saved” on the dashboard actually means
Last updated 12 September 2026
Two numbers sit next to each other and they are easy to mix up. One is how much smaller the model prompt got. The other is what you pay Hexum.
The send
Every MCP tool call records tokens_in: an estimate of what the
agent sent Hexum. Usage is €0.30 per million of that. Teams is €0.35 per
million. A 200 million month is €60 on Usage. There is no seat tax. Pricing.
The cut
When hexum_shrink returns a smaller view, the dashboard can
show how much was not forwarded to the model. That is “tokens saved.” It is
a measured cut on that turn. It is not a credit. It is not subtracted from
the invoice.
The published card is not the invoice either
Study I counts tokens sent to the model as ceil(utf8_bytes / 4).
Combined 220,393 → 27,352 (−87.6%), 24/24 pass. That protocol is on /benchmarks. When someone scales it to
$7,722 vs $1,094, the protocol has to travel with the figure: 10,000 ×
24-task suites, listed model prices, estimator — not an invoice.
Hexum does not hide the send. People will find it. Say it first, then show the cut.
We do not keep the code on Usage or Teams. Zero data retention.
FAQ
What does the dashboard mean by tokens saved?
How much smaller the reply was than what the agent sent, when that is a cut. Your invoice is still based on what the agent sent Hexum, not on the savings.
How is Hexum billed?
You pay for what your agent sent Hexum. Usage €0.30 per million tokens. Teams €0.35 per million. No per-person fee. No monthly cap. See /pricing.
How does Hexum cut what I pay the model?
Your agent stops dumping the whole repo into the prompt. Hexum sends this task, not every file. On the published card, MBPP context dropped 85.2% and the combined 24-task cut is 87.6%, with 24/24 tests still passing. Tokens are ceil(utf8_bytes/4), not a provider invoice. See /benchmarks.
Does Hexum store my repo?
No. We do not keep the code your agent sends. We do not train on it. Policy: /zdr.