Claude subscriptions can be 10-30x cheaper than the API. But you have no idea why you keep hitting limits.

I run Claude Code on a Max (x20) plan. Hundreds of sessions, thousands of turns. No breakdown by project, no model split, no cost visibility. Just a monthly bill.

So I built a local dashboard that reads Claude Code transcripts:

→ Cost per day, broken down by model

→ Sessions and turns over time

→ Filter by model, time range (7d / 30d / 90d)

→ All data stays on your machine

First thing I found: one day spiked to 700M cached tokens. An Anthropic bug.

30-day total: 440 sessions. 18,000 turns. $1,588 in estimated API costs.

All covered by the Max subscription. For a fraction of the price.

—

Install & run Windows (it will open localhost:8080):

git clone github.com/phuryn/claud…
cd claude-usage
python cli.py dashboard

—

Install & run MacOS (it will open localhost:8080):

git clone github.com/phuryn/claud…
cd claude-usage
python3 cli.py dashboard

—

Tracks any Claude subscription (Pro, Max) running Claude Code. SQLite for storage, Chart.js for charts, zero dependencies beyond Python.

Open source. MIT. GitHub: github.com/phuryn/claud…

—

On May 9, I'm launching Claudathon for PMs. You will ship full-stack products powered by AI agents. 100% hands-on. No-code.

More: go.productcompass.pm/cl…