AI usage & cost tracker · macOS + Linux
Every token.
Every dollar.
Every machine.
Tally runs quietly on each of your machines, reads the usage your AI tools already log, and syncs only the numbers to one dashboard: tokens, API-equivalent value, real spend and live plan limits.
Install in 30 seconds
- 01
curl -fsSL https://tally.arnavprabhu.com/install.sh | shInstall the agent (macOS · Linux) - 02
tally loginApprove this machine in your browser - 03
tally service installRun in the background, sync every 5 min
Single static binary · no root · open pricing table
- Claude$413.13
- Codex$288.48
- Cursor$97.83
- OpenRouter$60.45
What it tracks
Tokens
Input, output, cache read/write, reasoning — per hour, per model, per project.
API-equivalent value
What your usage would cost at public list prices, from a versioned pricing table.
Real spend
Subscriptions prorated per day plus actually-billed API dollars. The number your card sees.
Plan limits
Claude 5-hour & weekly windows, Codex, Cursor and more — live %, reset countdowns.
Every machine
One agent per laptop, server or VM. Buckets merge without double counting.
Every account
Personal, work, API keys. Rename or hide any account label.
Privacy promise
Numbers out.
Nothing else.
Prompts, code and keys never leave your machine. The agent uploads hourly totals — and you can read every byte with tally sync --dry-run.
↑ Leaves your machine
- Token counts
- Request counts
- Cost (USD)
- Model name
- Project folder name
- Hostname
- Hour bucket
- Plan limit %
- Account label
✕ Never leaves
- Prompts
- Completions
- Source code
- File paths
- API keys
- OAuth tokens
- Cookies