Skip to content

Usage & analytics

The Usage page combines provider state, billing configuration, and optional Hermes telemetry. Use the provider filter to show all connections or focus on one, and use the range control to choose the period being analyzed.

Overview

Four cards answer the quickest questions:

  • Cost combines configured subscription commitments with known PAYG cost bases. It warns when some PAYG cost is unavailable. Mixed currencies are not converted or added.
  • Tokens, Requests, and Sessions are Hermes-observed totals for the selected range (with compatible provider activity as a fallback for tokens or requests where available).

When one provider is selected, the cards are scoped to it. If several connections share that provider, Hermes workload cannot be assigned safely to one configuration; the page asks you to select All providers for the shared provider-level total.

For subscriptions, the Overview displays the full configured commitment, not the backend's selected-range allocation. See Understanding cost & value.

Provider usage & quota

Each connection card shows billing type, observed tokens and requests, cost, and every provider-native quota window the adapter can normalize. Every window has its own used percentage, status, optional remaining percentage, provenance, and reset time.

PAYG connections without quotas show the selected spend source. Subscription providers without a quota say so rather than displaying 0%. Stale or unavailable provider readings are labelled on the card.

See Understanding quotas and Provider health & errors.

Usage over time

This chart shows daily Hermes-observed workload. Choose tokens, observed cost, requests, or sessions, and group by provider or model.

Observed cost here is Hermes telemetry. It is not subscription commitment, provider-reported spend, or the Cost & Value cost basis. Missing dates are gaps; a reported zero is shown as an observed zero. These states are deliberately not interchangeable.

Cost & value

The table shows each provider's billing model, displayed cost, Hermes-observed tokens and requests, Tokens/$, Requests/$, Cost/1M tokens, and share of observed tokens. Hover the cost to see its source, pricing coverage, and catalogue version.

For PAYG, displayed cost follows the safe source precedence. For subscriptions, the active page displays and uses the full configured commitment for these table ratios. The economics API separately calculates a real-period allocation in cost_basis; that allocation is not currently surfaced by this table. Compare ranges with this distinction in mind.

The line below the table reports the API-equivalent estimate and global Hermes pricing coverage. An unavailable result means no workload could be priced safely; a partial result identifies incomplete coverage.

Breakdown

Tabs group Hermes-observed cost, tokens, requests, sessions, and token share by:

  • provider;
  • model; or
  • Hermes profile.

These totals describe only workload observed by Hermes. They are not added to provider-native totals. See Hermes attribution.

Data sources & quality

This collapsed section summarizes healthy Hermes sources, stale/unavailable provider snapshots, pricing coverage, and unpriced models. Expand it to see each source's health, latest observation, number of observations in range, unresolved provider aliases, pricing catalogue version, unpriced token count, and diagnostic messages.

For deeper API-level information—including reconciliation, confidence, forecast, capacity, and attribution endpoints not currently rendered in this page—see Analytics configuration and behavior and the API reference.

Self-hosted API usage dashboard.