See where your LLM usage money actually goes.

• Tokens are tokens. input_tokens, outputTokens, cache_read_tokens: anything named for tokens is counted as tokens, and totals read as 22.6K tokens, not 22565.
• Cost is money. Fields named cost, price, spend or fee are dollars, or rupees, euros and others when the name says so (cost_inr, price_eur).
• Models and purposes are categories. Breakdowns and charts by model, feature or endpoint come out on their own.
• Timestamps are time. ISO dates and Unix timestamps become a timeline of calls and spend.
• JSON or JSON Lines. A summary file with an array of calls, or a JSONL log with one call per line.

• Totals that don't match. If the log's summary says 22,565 input tokens, the calls are added up and checked against it.
• Where the money goes. A few calls, models or purposes carrying most of the spend.
• Outlier calls. Prompts far larger than usual, often a bug that sends a whole document twice.
• Duplicates. The same call logged twice, as retries often are.
• Evidence attached. Every finding lists the JSONPaths of the values behind it, so you can check it in a click.

LLM usage logs FAQ

Which LLM usage logs can I open?

Any log written as JSON or JSON Lines: a usage export from a provider's console, a gateway or proxy log, or the records your own app writes for each call. There's no fixed format to match. Fields are recognised by their names, so input_tokens, outputTokens and cost_usd all work.

How does it know which fields are tokens and cost?

From the field names. A field with "tokens" in its name is counted in tokens. One named cost, price, spend or fee is money, in US dollars unless the name says otherwise, as in cost_inr or price_eur. Timestamps, model names and IDs are recognised too.

Are my prompts uploaded anywhere?

No. The log is read and analysed in your browser and stays on your machine. That matters for usage logs, which often contain prompts, responses and customer data.

Can I compare usage across projects or days?

Yes. Choose several log files or a folder in the app. Files with the same structure are stacked into one dataset with a column saying which file each call came from, so you can compare projects, days or environments side by side.

What does it flag automatically?

Summary totals that don't match the calls they summarise, calls far above the usual tokens or cost, a few calls or models carrying most of the spend, duplicate records such as retries logged twice, and gaps in time. Each finding links to the exact values behind it.

Is it free?

Yes, at any file size and with no account.