SaaSpec

Logging takes a single HTTP call to SaaSpec's endpoint with five fields: user ID, input, output, model name, and your API token. There's no SDK to install and no agent to run on your servers. Dropping the call into an existing handler takes a few lines, which matters when you're adding billing to an app that already ships.

Every record is filed under the UID you pass in. That string has to line up with how the same person appears in Stripe, or with something you can map back to an email when you invoice. Get that key wrong at the start and the monthly export becomes a list of costs with no owner, so the naming decision deserves a minute of thought before launch.

SaaSpec asks you to send the model name exactly as the API returned it, not a hardcoded string. Models get renamed and repriced, and a static value in your code will quietly drift from reality. Passing the response value keeps the pricing table in sync with whatever model actually served the request.

The export includes an anomaly column alongside cost, input count, and output count. It marks accounts whose usage sits far from the rest, which is the fastest way to spot a runaway script, a leaked key, or a customer who found an expensive prompt. You still have to act on the flag, but you don't have to build the detection yourself.

One export per billing period gives you a table you can open in a spreadsheet, hand to an accountant, or parse in code. Each row pairs a user with their token cost and their input and output counts, so a support question about a charge can be answered from the same file.

Stripe invoicing isn't built in, but the docs hand you a working pattern: read the CSV, find or create a Stripe customer by email, add an invoice item for the token cost, then create and send the invoice. Paste it into a script, adjust the column names to your export, and run it once a month.