Invoices from your Claude Code sessions

If most of your billable work now happens in a conversation with an AI, the trail it leaves is better evidence than any timer. Here is how to bill from it.

Billing got strange when the work moved into AI sessions. A day where you shipped three features can look like almost nothing in a commit log, because the commits are large and few, and it looks like nothing at all in a timer you never started. Meanwhile the actual record of what you did, the session itself, sits on your disk being ignored.

Why AI sessions are unusually good evidence

A timer records that you were present. A commit records that something landed. A Claude Code session records the thing in between, which is the part that is hardest to reconstruct three weeks later:

• What you were trying to do, stated in your own words at the time
• Which files and folders it touched, which is what identifies the client
• When it happened, to the minute
• The exploration that produced no commit, the approach you tried and abandoned, which is real billable work that git alone never sees

That last one matters more than people expect. A morning spent working out that an approach will not work is billable, and it is invisible to every commit-based tool.

How Get Thy Bread reads them

It runs in the macOS menu bar and reads your Claude Code and Cursor sessions locally, alongside the git activity in the folders you have assigned to each client. Sessions are routed to the right client by the files and repositories they touched rather than by guesswork, so work in a client's folder bills to that client and your own side projects stay unbilled.

Once a day a popover shows what it saw, collapsed from a hundred micro-steps into one honest summary. You confirm it in a tap, correct it, or add the meeting it could not see. Anything it is unsure about is flagged rather than turned into hours.

By the end of the month there is nothing to reconstruct, because every day was confirmed while you still remembered it. Details of the capture are in how automatic time tracking works.

The question everyone actually asks

If AI made you three times faster, should the client pay a third as much?

This is a pricing question wearing a tracking question's clothes, and it is worth separating the two. What you bill is a decision about value, leverage, and what you agreed. What you track should simply be true: the hours you actually spent, on what, for whom.

The failure mode worth avoiding is letting the ambiguity turn into under-billing by accident, which is what happens when you reconstruct a blurry month from memory and round everything down. An accurate record lets you make the pricing decision deliberately instead of losing the argument to your own forgetfulness.

What it does not do

• It does not watch your screen or take screenshots. It reads artifacts, not activity.
• It does not upload your session history. Capture is local to your Mac.
• It has no team dashboard and no manager login, so no one can be given a view of your day.
• It cannot see work that leaves no trail. Meetings and calls you still add by hand, in a sentence.

If you also use Cursor

Cursor sessions are read the same way and land on the same timesheet. A day split between Claude Code, Cursor, and ordinary commits produces one set of billable lines rather than three records you have to reconcile.

Comparisons, if you are weighing this against what you use now: Toggl Track, Harvest, FreshBooks.

Common questions

How do I track billable hours when I work in Claude Code?

Your sessions already record what you built and when, which is stronger evidence than a timer you forgot to start. Get Thy Bread reads those sessions locally on your Mac, works out which client the work belongs to from the folders and files involved, and drafts each day for you to confirm.

Should I charge less because Claude wrote some of the code?

That is a pricing decision, not a tracking one, and most freelancers land on billing for the outcome rather than the keystrokes. If a feature that used to take two days now takes four hours, you have either delivered the same value faster or you can take on more work. The honest position is to bill for the time you actually spent, which is exactly what evidence-based capture gives you a record of.

Does it read the contents of my Claude Code conversations?

It reads your local session records to work out what you were doing and which files it touched, all on your own Mac. To write the one-line summary of a day, the summarised work items are sent to a hosted model. Your raw session history is not uploaded, and there is no team dashboard or manager view that could expose it to anyone else.

Does it work with Cursor too?

Yes. Cursor sessions are read the same way and land on the same timesheet, so a day split between Claude Code and Cursor still produces one coherent set of billable lines.