See where your AI money goes.

Drop in a week of LLM logs. You'll get your monthly spend by workflow and model, the biggest leaks, and what fixing them is worth in dollars.

Drop your log export here

JSONL, JSON or CSV from Langfuse, Helicone, LiteLLM, Portkey or your own logging. It's read in this tab and never uploaded.

Leave on auto if your logs have timestamps.

Prefer to do it together? Email hello@thetechguys.dev and an engineer will run the audit with you on a call.

What should the file contain?

One row per LLM call. The audit recognises common field names from most tools. The more of these you include, the more it finds:

  • Required: model, input tokens, output tokens
  • Strongly recommended: workflow or feature name, timestamp
  • Helpful: trace or session ID, cached tokens, cost, status or error, the prompt (used only to spot repeats)
  • Sampled logs: add a weight field for how many calls each row represents

Example JSONL row:

{"timestamp":"2026-10-28T09:14:02Z",
 "workflow":"ticket-triage",
 "model":"claude-opus-5",
 "input_tokens":1412,"output_tokens":9,
 "trace_id":"t_8f2c","status":"ok"}

Prices are approximate list prices. When your logs include a cost field, the audit uses it instead.