written by the modelnew text readre-read from cache
Each row is one thing you asked. Click a row to open its model calls, then click a call to see what it sent and what it wrote. The parts always add up to the number above them.
How to read: one row per message you sent. The colored bar is its cost: longer means more expensive, and the colors show what the money paid for (purple: the AI writing, green: new text read, amber: re-reading the conversation). Click a row to see the model calls behind it.
Replay
Walk the session call by call. The cost meter climbs; the second bar shows how much text was sent with each call (the agent's memory at that moment).
How to read: press Play or drag the slider. Each bar is one model call, in time order; taller means it cost more, and blue bars have already happened. The strip under the bars shows which of your messages each call belongs to; click a turn to jump there. The two meters show the money spent so far and how much text this call sent and wrote.
Tokens sent so far
Tokens in / out for this call
Where the tokens went
By kind of step, adding up to the whole session
How to read: each bar is one kind of work the AI did, biggest first. Every bar starts where the one above it ended, so they step to the right and finish at the dark bar, the whole session. A longer bar means more money; the numbers on the right are its cost and tokens. Hover a bar for details, click it to list its calls.
By kind of token
How to read: each bar is one kind of token (text) the AI paid for. A longer bar means more of the cost; the right side shows how much and its share.
What to fix
tokens cut, measured from this session
Each saving is what those tokens or calls cost in this session, priced like the rest of this page. Fixes can overlap.
Sent with every call
Helpers
Helpers (subagents) work in their own chat with their own memory. Their cost is included in the total.