Your bill
Reads a usage log and says where the money actually went: which workload, which model, whether caching paid for itself, and which levers would move the bill. Read entirely in this browser tab.
Nothing is generated and nothing is sent to a model. The questions are fixed, and the words in the prompt are yours.
Nothing priced yet
Paste your prompt and press Optimise to see what can go and what it costs.
- what the rules can remove, with the diff
- what it costs a month, and what the saving is worth
- what else is worth fixing: caching, model tier, the Batch API
Nothing compared yet
Two versions of the same prompt. What the edit did to the token count, what that costs, and which problems it introduced or resolved.
- the token count before and after, and what the difference costs
- the monthly figure at your call volume, signed either way
- which advisories the edit introduced, and which it resolved
The log
One JSON object per line, each with a "model" and the "usage" object the API returned. A "label" names the workload and a "session" names the conversation; both are optional, and both unlock findings the log cannot answer without them.
Recording the log is three lines in your own code: after each API call, append one JSON line with the model, the usage object the response carried, a "label" naming the workload and a "session" naming the conversation. It never contains prompt text, and the session key is grouped by and never shown.
Files stay in this browser tab. Nothing is uploaded or saved by Trazum. Clear the workspace or close the page to discard them.
Every repeated call needs context.
Choose a recorded run and describe the acceptance check. The report will separate recorded tokens, repeated results and gaps in the evidence.
Read the task workflow ↗Type "help" to see what runs here. Nothing is uploaded and nothing is fetched.