What Cursor really puts into each project

• Every agent session from start to end, with the model (Cursor 1.7 or later).
• A title per session: a short title from the first prompt, later the chat name Cursor assigns.
• The folders the session wrote to, for assigning it to the right project.
• Tokens and cost, fetched afterwards from your Cursor account (see below).
• Commits come in through the GitHub connection.

Cursor's hooks report no tokens. The collector fetches them from cursor.com, the way Cursor's own usage dashboard shows them.

• It uses your local Cursor sign-in read-only. That credential only goes to cursor.com, never to Time Momentum.
• Usage is matched to a session by time. If two agents run at the same moment with the same model, part of it can land on the wrong session; the total is right.
• Cursor's interface is unofficial. If it changes, sessions keep being captured with their duration, just without tokens. You can switch the fetch off.
• There is no measured active runtime and no import of past sessions, because Cursor keeps no transcript to read.

Models with a known list price are priced like in every other tool, Cursor's own models with the list price Cursor reports. On the invoice, Cursor work appears as cost with a markup or as a token quantity at your price; billing by runtime needs a measured active runtime, which Cursor does not provide. Passing AI costs on is part of Pro.

Prompts, responses and code never leave your machine. Time Momentum receives the model, tokens, runtime, working folder, repository and a session title. Where a tool keeps no title, the first prompt is sent shortened to 200 characters; Time Momentum turns it into a short title and does not store the prompt. You can switch off titles and folders entirely or exclude individual folders. The capture code is public in the npm package @time-momentum/cli.