clauditor
Claude.ai shows you a usage bar. Clauditor shows you everything else.
Which project burned through the most tokens this week? Which session cost $12 in one sitting? How much are you actually saving from prompt caching? Your Claude Code subscription gives you none of this — clauditor does.
It parses the JSONL transcripts that Claude Code writes locally, stores them in a SQLite database on your machine, and gives you a terminal UI and web dashboard to explore your usage. No API key. No account. Completely offline.
Built for Claude Code Pro and Max subscribers who run it all day across multiple projects and want to know where their usage is actually going.
What you get that Claude.ai doesn't
|
Claude.ai |
clauditor |
| Usage bar (% of limit) |
✓ |
✓ |
| Cost per session |
✗ |
✓ |
| Breakdown by project |
✗ |
✓ |
| Daily usage trends |
✗ |
✓ |
| Cache read vs creation savings |
✗ |
✓ |
| API-equivalent cost estimates |
✗ |
✓ |
| Model-by-model breakdown |
✗ |
✓ |
| Works offline, no account |
✗ |
✓ |
Installation
With Go 1.25+
go install github.com/sadhakbj/clauditor@latest
This installs the binary to $GOPATH/bin (usually ~/go/bin). Run the same command to update to the latest version. Make sure that's on your $PATH:
export PATH="$PATH:$(go env GOPATH)/bin"
Check the installed version with:
clauditor --version
Without Go — download a pre-built binary for your platform from GitHub Releases and put it somewhere on your $PATH.
Quick start
No setup required. The database (~/.claude/usage.db) is created automatically on first run.
# Scan your transcripts and open the dashboard
clauditor dashboard
# Or use the terminal UI
clauditor tui
# Just check today's usage
clauditor today
Usage
clauditor [command] [flags]
Commands
| Command |
Description |
scan |
Parse JSONL transcripts and write to the database |
today |
Print today's usage broken down by model |
stats |
Print all-time statistics |
dashboard |
Scan + start a local web dashboard |
tui |
Launch the interactive terminal UI |
Global flags
| Flag |
Default |
Description |
--db |
~/.claude/usage.db |
Path to the SQLite database |
--dir |
~/.claude/projects |
Path to Claude projects directory |
Dashboard flags
| Flag |
Default |
Description |
--port |
8080 |
Port for the web dashboard |
--no-browser |
false |
Don't open the browser automatically |
Web dashboard
Runs at http://localhost:8080. Shows:
- KPIs: sessions, turns, input/output tokens, cache usage, estimated cost
- Daily token usage (stacked: input / output / cache read / cache creation)
- Breakdown by model
- Top projects by cost
- Session list and request-level drilldowns
Filter by model and date range (today / 7d / 30d / 90d / all time). Refreshes automatically every 60 seconds — no page reload.



Terminal UI
clauditor tui

| Key |
Action |
1 / 2 / 3 |
Overview / Sessions / Models |
tab |
Cycle views |
t / d / w / m / a |
Date range: today / 7d / 30d / 90d / all |
j / k or ↑ / ↓ |
Scroll / navigate |
r |
Re-scan and refresh |
/ |
Filter sessions |
? |
Help |
q |
Quit |
How it works
Claude Code writes every conversation as a JSONL file under ~/.claude/projects/. Each line is a turn — it includes the model used, input tokens, output tokens, and cache token counts. Clauditor reads those files, aggregates them into a local SQLite database, and renders the results.
Nothing leaves your machine.
Cost estimates
Estimated using Anthropic API pricing (April 2026). Your subscription cost is different — these numbers show what the same usage would cost on the pay-as-you-go API, which is useful if you're evaluating whether Claude Code is worth it for your team, or comparing projects by spend.
| Model |
Input |
Output |
Cache write |
Cache read |
| Opus 4.x |
$6.15/M |
$30.75/M |
$7.69/M |
$0.61/M |
| Sonnet 4.x |
$3.69/M |
$18.45/M |
$4.61/M |
$0.37/M |
| Haiku 4.x |
$1.23/M |
$6.15/M |
$1.54/M |
$0.12/M |
Tech