SessionCost

Turn by turn

0 · fixed setup1 · earlier conversation2 · newre-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 bar is the tokens it sent, in three parts: grey, the agent's fixed setup (step 0, sent with every call); blue, the earlier conversation sent again (step 1); green, what was new (step 2). The thin amber line under it is how much of that was re-read from the cache, cheaply; the rest went at full price. 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.

Deeper fixes

These are the three basic checks. Our catalog of measured waste patterns has dozens more: an agent polling instead of waiting, files re-read, tool results far bigger than the work, research in the main chat, and others. To check them, SessionCost would share a summary of this session: numbers, tool names and yes/no flags. Never your messages, code, file names or project.

Sent with every call