Usage and spending

Read request cost, token use and reliability without confusing them with wallet credit.

Open Analytics to understand activity in the current organization. Start by checking the organization, audience and time range.

Home is a present-tense summary for every member, not a report: your budget or, for an owner who can see billing, the wallet runway, hours and live sessions, and what needs you. Organization totals, rankings and member investigations live here in Analytics. Workload classes are reported categories, not workflow traces.

Choose an audience

Me shows your requests, narrowed to the current organization in the customer app.
Team shows an allowed team's attributed requests. Membership alone does not imply access to every team's usage.
Organization shows aggregates when your permissions allow it.

Choose a time range (Today, This week, a preset or a custom period shown in UTC; future dates are refused), then use the Overview, Costs, Reliability and Requests tabs. The Advanced detail level adds Breakdown and Value tabs, previous-period comparison and token composition. Moving between views should keep the context you selected. A missing team, failed team lookup or mismatched organization bookmark is a recovery state, not an empty result to interpret as zero spend.

Organization usage administrators can search the All members picker to filter by a member, including members with no usage in the selected period. Member names come from the organization's spending targets independently of usage results, so they remain readable when changing dates or reopening a link. Links retain stable member IDs; hover over a filter chip to see its full label and ID. If a name cannot be resolved, the filter stays applied with an explicit unavailable label. Retry a failed name lookup or choose All members to remove the member filter.

Compare members in the leaderboard

Open Analytics → Leaderboard with Organization scope. Organization usage administrators, including delegated billing administrators, can rank members, teams, models, providers, API keys, workload classes and tags by Usage cost, Requests or Tokens. Changing the ranking metric updates both bars and table ordering. Shares use the filtered total; the table includes comparisons against the preceding period. Comparison rows outside the previous top 100 are marked unknown unless a complete previous result establishes that there was no activity.

The member picker searches the organization's spending targets. When the complete current ranking fits on one page, a No usage in this view section also lists members absent from that result. A partial page or a failed query never establishes zero usage. Ranking is consumption analysis, not a measure of employee productivity.

Inspect a member

Choose a member from the leaderboard. The dedicated /analytics/members/<memberId> page retains the organization and period. The member stays fixed while dates, tabs and other filters change; Back to leaderboard returns to the comparison context.

Overview shows totals, changes and activity. Choose a busiest-bucket button to open its precise UTC request window.
Cost & tokens groups activity by model, provider, key, team, agent, tag or workload class. Select a supported breakdown row to open filtered requests. Workload classes are categories and cannot narrow rollup-backed request views.
Efficiency shows usage cost/request, tokens/request, reliability and reported token composition. Cache counters may overlap input counters; no universal cache-hit percentage or hypothetical savings is inferred.
Requests supports searching loaded records by model, provider, key, execution ID or error. Load more to include older records in that window. Expand a request for available recorded execution and settlement evidence.
Results shows reported outcomes and their attribution; Clients separates routed client usage from supplemental client reports.
Budgets & policies shows current organization/member policies and contextual team policies. Request-bound teams determine applicability; these are not historical usage-limit verdicts. Monetary limits retain their exact values.

Member usage may include automation under their keys. Compare similar workloads and models and consider sample sizes and outcome attribution. Usage permission does not reveal private profile fields, prompts or API-key secrets. A member link requires the intended organization and active-organization usage authority.

Three different figures

Figure	Answers
Usage cost	What cost was reported for the selected requests and time range?
Budget remaining	How much capacity is left under the applicable usage-limit policy and window?
Available credit	What can the organization's wallet fund now, after reservations?

Reported request cost is not proof of a settled wallet debit. Billing can be disabled or observing; requests may still be settling. Provider catalog pricing is explanatory data, not a wallet transaction. Use Billing & credits for credit and settled wallet totals.

Budget capacity and wallet credit are separate checks

Usage budgetDoes the policy allow more usage in this window?A blocking limit can return 429.

Available creditCan the payer fund this request?A credit refusal can return 402.

Buying credit does not increase a budget. Raising a budget does not add credit.

A request must satisfy every applicable blocking budget and, when billing is enforced, credit admission. Other access and model checks still apply.

Agents

Agents groups API usage by the agent type assigned to each key: Claude Code, Codex, OpenCode, a Custom name you supply, or Unassigned for keys without an assignment. Assign the type when creating a key, or change it later from API keys in the account menu if you own the key. Usage keeps the assignment the key had when each request was admitted, so a change affects future requests and earlier rows stay where they were recorded. Point one tool at one key; several tools on the same key are indistinguishable.

Your own keys are always visible. The organization scope, linked-key counts and configured agents with no calls need organization usage authority, and other owners' key names appear only to organization administrators. Agent views cover periods of up to 92 days.

Investigate a request

Open its details to inspect the model, usage, timestamps and available settlement information. Preserve the execution/request identifier when asking your operator for help. Do not include API keys or sensitive prompts in support messages.

When an execution ID is available, expanding the row loads its Recorded attempt & settlement evidence. The view separates customer charge, rated usage cost and the authorization ceiling. A ceiling is not a charge, HTTP success is not settlement, and neither establishes a successful business outcome. Missing evidence remains unavailable.

In the organization Requests view, administrators can Find an execution or related attempts using an execution UUID or client request ID. This lookup searches the active organization independently of the chart's date and member/team/model filters. Client request IDs can be reused: matches do not prove retries or a complete workflow. The service returns at most 50 newest matches here and warns when more may exist.

The Value view shows reported results and their attribution. Expand a result's reported execution IDs to follow one into evidence when organization usage permissions allow it. The reporting application supplies these links; they are not productivity scores.

Frozen pricing and settlement-policy versions describe billing policy at admission. Current policies on Home and Budgets & limits show today's configuration; they do not establish which usage-limit decision applied to a past request. Historical usage-limit decisions and intermediate provider retries are not included in this evidence.

An empty view may reflect the audience, period or filters. A failed load should offer an error and recovery action; do not treat it as zero. Usage collection and settlement can complete asynchronously.

Older Billing Spend bookmarks lead to this workspace with their caller audience and date range preserved. FOCUS export is outside this customer onboarding guide's verified scope.

A request that named tokamak/auto shows the model that served it with an auto badge. Its details say which tier served and why. The Auto routing tab (Advanced view) sums those decisions; see Read the routing report.

See Budgets and limits, Credits and billing and the Usage API.

Understand client adoption in Tools

The Tools tab compares routed coding-client activity within the current organization and filters, over at most 92 days. It shows request and token share, active credential owners and keys, requests per elapsed minute, exact usage cost, cost per request, errors, 429s and latency. Unknown clients remain in totals; coverage shows how much evidence is labelled and has status/token information. An owner is a credential owner, not necessarily one human. Owners can use several tools, so adoption percentages overlap. Empty evidence is unavailable, not zero.

Client logs, metric samples and execution spans appear separately. These reports are supplemental and may be missing; they never add another usage charge. A metric sample is not a task or an extra billable request. Provider filters hide client activity because it cannot prove which routed provider handled a request. General historical views may retain old OTLP imports; Tools excludes those imports.

Use these measures to investigate consumption and reliability, then apply existing organization, team, user or key controls. They are not productivity scores or seat adoption rates. Usage cost and settled usage cost are not wallet debits. Client labels cannot grant authority or change price. See Budgets and limits for actual controls.