The problem
I've been using Claude Code heavily for the past few months.
But I had absolutely no idea how many tokens I was actually burning — or where.
No visibility. No breakdown. Just vibes.
What I found
Turns out, all the data is sitting right there on your machine:
~/.claude/projects/
Every session is logged as a JSONL file with full token counts —
input, output, cache reads, cache creation.
Nobody told me this existed.
What I built
A simple Next.js dashboard that reads those files locally and shows:
- Token usage per project
- Daily usage chart
- Cache read ratio
When I ran it on my own data, I discovered my cache read ratio was 95.8%.
I had no idea that was happening on every single turn.
My heaviest day: 340,000 tokens in a single day on one project.
I had no memory of it being that intense.
The insight
This isn't just about cost.
It's about visibility.
When you can't see what's happening, you can't improve it.
The data was always there — I just couldn't see it.
Try it yourself
GitHub: https://github.com/notenkitoclient-cpu/claude-token-dashboard
Clone it, run it locally, and find out what's actually happening in your Claude Code sessions.
Would love to hear what numbers you find.