I will audit your claude code token usage and cut the waste


About this gig
Your AI agent is burning tokens and you don't know where.
Common advice: smaller model, shorter prompts. I measured a real setup - 551 sessions, 22.6B tokens in 14 days. Both were the wrong place to look.
What was actually true:
- 98.8% of the spend was cache_read: re-paying for the same context, over and over.
- Cost = requests x context size.
- Scheduled automation was 12% of the bill. Two long sessions, 47%: one ran 45 hours, 24,111 requests.
- Loading MCP servers: nothing. Trimming 186 unused skills: 1%. Both are noise.
The money sits in one or two places, rarely where you'd guess. I read the logs your tool wrote.
WHAT YOU GET
- Automated runs vs a human at the keyboard
- Most expensive sessions: duration, requests, cost
- Which tool is filling your context
- Your context floor: what every request pays before you type
- Three ranked fixes, sized from your own data
YOUR DATA STAYS YOURS
You run one local script and send me a summary file: numbers only, no paths, no prompts. I never touch your machine.
MONEY-BACK GUARANTEE - FIRST 10 ORDERS
If it doesn't show where your spend goes and at least one change worth making, I refund in full.
Get to know Fernando Rodas
AI assistants and automations that answer, qualify and quote around the clock
- FromGuatemala
- Member sinceAug 2023
- Avg. response time2 hours
Languages
Spanish, English
My Portfolio
Other AI Development Services I Offer
FAQ
Do you need access to my computer or my account?
No. You run one local script and send me the summary file it writes. I never touch your machine, your code or your prompts.
What if I am not happy with the report?
Then you do not pay for it. On the first ten orders I refund in full, no questions and no forms. I would rather lose one order than sell you a report you cannot act on.
Does this work if I am on a subscription plan and not paying per token?
Yes, and it is often more useful. The dollar figures become a way to rank problems by size. What you get back is the afternoon you keep losing when the quota runs out.
What if my setup is not Claude Code?
Message me first. The method reads session transcripts, so it ports to any harness that writes them. I will tell you honestly if yours does not.
Will this break anything?
The Report package changes nothing at all, it only reads. The Meter and Full packages add a counter and a spending ceiling. Both are reversible and I hand you the rollback.
How much will I save?
I will not promise a number before seeing your data, and be careful with anyone who does. What I promise is that you will know where it goes and what to do about it.
