Insights¶
Charts and numbers: token and cost usage over time, per agent and per model, thread counts, and activity trends. If your deployment displays costs in credits rather than dollars, this section follows that setting.
The figures cover everything your deployment runs, not just chat: Cortex background thinking, scheduled tasks, delegated skill workers, the daily highlights and periodic reports, post-turn PULSE curation, subagent calls, and the non-chat services (image generation and editing, vision, voice, web search) all count. A "Where credits went" card breaks the window down by kind of work — chat and channels, Cortex, scheduled tasks, and so on — which is the place to look when usage seems higher than your conversations alone explain, and a "Credits by tier" card splits managed model spend across the Turbo / Pro / Max tiers so you can see which tier is doing the work.
Every number here is measured on your own deployment: each model call and service call records its charge locally as it happens, so the charts, the breakdown cards, and the headline credits figure all come from the same local record and stay available even when the deployment is offline. That one record also backs every range button, so the ranges nest the way you would expect — 1 day is a rolling last-24-hours view and is always contained in 7 days, which is contained in 30 days and 90 days. A shorter range can never show more calls or more spend than a longer one; if it ever does, something is wrong and it is worth reporting. The range buttons mean the same thing on the Drill-down tab, which shares the selector — "1 day" is the last 24 hours there too, scoped to the conversations active in that time.
Drill-down's agent ranking and its chat-vs-autonomous cards read the same record as the Overview, so an agent's total is the same number on both tabs. What changes is when you open an agent: from there down, the figures are attributed from stored conversation history, the only record that ties a cost to a particular conversation or tool. Long conversations are compacted as they grow, which removes older calls from that history, so the conversations you can open account for part of an agent's total rather than all of it. The page says how much — "openable conversations account for $X of $Y" — so a short list under a large total reads as coverage rather than as a contradiction. One more difference follows from what it ranks: a conversation is counted whole, in the range its most recent activity falls into, so a long-running thread brings its entire cost with it. Take the Overview and the agent ranking as the totals, and the levels below as the answer to where inside a conversation the cost went.
Billing itself still lives with your Olano account — the Credits left balance shown on the Overview page and the usage bar on your cloud deployment page track the account's remaining balance for the billing period, which is a different window than the charts' last-N-days view, so the two totals are not expected to match digit for digit.
On deployments upgraded from an older version, the dashboard reconstructs past Cortex background spend from its stored history the first time it starts, so the charts show the full picture rather than beginning at the upgrade day. Reconstructed history is marked internally and never re-billed — your balance is unaffected.