Understand savings

Run cmx stats to see local usage. If you linked a CMX account, cmx dashboard opens account-linked aggregate statistics. The dashboard receives aggregate counters, not individual requests or session content.

Figure	What it means
Tokens handled	Token activity observed by the gateway; this is not a count of saved tokens.
Estimated saved tokens	An estimate of input tokens avoided by accepted compression, with internal summarization usage accounted for separately.
Cache reads	Input tokens that the model provider reported as read from its cache. They are separate from CMX’s estimated compression savings.
Internal usage	Tokens spent on CMX’s summarization calls. This cost can outweigh a short session’s savings.

These figures are not a provider invoice. Cached input may cost less than ordinary input, models have different output prices, and providers can report usage differently. Use your model provider’s billing page for actual charges.

Check whether CMX is receiving traffic

Run cmx stats and note the request count.
Send a normal request in a connected coding client.
Run cmx stats again.

A new request is evidence that the client reached CMX. It may show zero compression in a short session or when policy found no safe, economical span. If nothing changes, check cmx harness verify, restart the client, and follow Troubleshooting.

Last modified on September 24, 2026