OFFLINETL-01
Offline-first
Your logs never leave this machine. No account, no telemetry, no network call.
TokensLab
The cost lab for your AI
Watch every dollar you spend on Claude, Cursor and Codex.
Offline. Private. Free to download.
Monitor AI spend like a professional
- Naive arithmetic
- $15,680.35
- Cache-aware
- $3,123.97
- Left on the table
- $12,556.38
Measured on 66 MB of real Claude Code session logs.
compat
The agents on your machine
Claude Code is live. Cursor, Codex and Gemini are on the tray. Copilot, Cline, Windsurf and the rest of this bench are the compatibility map.
01 / install
Download TokensLab
We detected you are on macOS
Download TokensLab for macOSOr pick your system
SHA256
release checksums on GitHub
View checksums
02 / specimen tray
Why TokensLab?
PRICINGTL-02
Cache-aware pricing
Cache-read tokens are billed at 0.1×. Naive math overstates the bill by ~5×.
SKILLSTL-03
Skills Explorer
See which prompts, tools and context packs actually show up in your sessions.
GRAPHTL-04
Graph Explorer
A tree of files your AI touched, sized by what they cost.
ALERTSTL-05
Idle session alerts
Sessions that went quiet still cost money. Get a notification before they linger.
INSTALLTL-06
Installer, not a clone
Download a build for macOS, Windows or Linux. GitHub is for stars and releases — source stays private at launch.
03 / lab bench
See TokensLab at work
Dashboard with the numbers that matter
04 / rate card
Free + Enterprise
Free
- Free desktop app
- Official installer
- macOS, Windows, Linux
- Offline-first
- Free forever
Enterprise
- Everything in Free
- Team workspaces
- SSO
- Audit logs
- Priority support
Every feature in the app is free. Enterprise is for teams that need shared workspaces.
06 / protocol
Frequently asked questions
What is TokensLab?
A desktop app that reads AI coding-tool session logs from disk, prices every turn against the published rate card, and shows the result: spend, models, projects, live sessions and how fast you are burning money.
Do my logs go to the cloud?
No. Everything is processed locally. There is no account, no telemetry and no network call from the app.
Is it really free?
Yes. Download the installer. No account. The source is not public at launch.
How does cache-aware pricing work?
Claude Code records five token classes, not one. Cache reads bill at 0.1× the base input rate, cache writes at 1.25× or 2×. TokensLab prices each class separately so the total is a real invoice, not a 5× overestimate.
Which operating systems are supported?
macOS, Windows and Linux. Download the installer from this site or GitHub Releases.
Is the source open?
Not at launch. Star the GitHub repo for releases. The desktop app still runs fully offline.
05 / observations
From the lab notebook