Usage tracking
See how many tokens your AI agents use
The aicharts collector counts tokens, cost, and speed for each model across the coding agents on your machine. Prompts, transcripts, file paths, and provider credentials stay on your machine, and each total lists the sources it covers.
In development. Apple silicon Mac and Linux x86-64 builds are on GitHub Releases; on an Intel Mac, build from source. Account sync runs on macOS only.
- macOSApple silicon
- Linuxx86_64, local reports
Similar tools
ccusage and Tokscale also read coding agents’ local logs. ccusage runs without an install and prints daily, weekly, monthly, and session reports. Tokscale adds a web dashboard and a public leaderboard, and the aicharts collector builds on its open-source parsers. aicharts shows which sources each total covers, and on a Mac it can sync daily totals to your dashboard. Checked Sep 28, 2026.
Set up tracking in three steps
The collector runs on your machine. When it publishes, it uploads daily totals of tokens, cost, and time for each client and model, with each client’s source coverage.
Install the collector
Choose your platform below. The collector reads token counts from your clients' local files. A few clients, such as Cursor and Warp, need a refresh step first.
Connect your account
On a Mac, run aicharts enroll and approve the pairing in your browser. Each enrolled Mac reports to one account.
Publish on a schedule
Set up a launchd job that runs aicharts autosubmit. Each run refreshes your clients and uploads their latest totals, so your dashboard stays current without manual exports.
macOS
curl -fsSL https://aicharts.io/install.sh | shLinux
curl -fsSL https://aicharts.io/install.sh | shWindows
aicharts enroll --state-dir "$HOME/.aicharts/state"What the dashboard shows
Every figure comes from the records your clients keep locally, reported per client, model, and UTC day.
- Tokens
Token types
Input, cache read, cache write, output, and reasoning tokens are counted separately.
- Cost
Reported vs retail
Provider-reported spend beside an estimate at public prices from the models.dev catalog, with the difference between them.
- Speed
Tokens per second
Measured over the request durations that clients record, per model and per day. This is not the model's inference speed.
- Cache
Cache reads
The share of input tokens served from cache, so you can see where prompt caching helps.
- Coverage
Source coverage
Each report lists every client it checked and whether it had data, was empty, was not found, or could not be read in full, so missing sources stay visible.
- Clients
55 sources
Codex, Claude Code, Cursor, Devin, OpenCode, Warp, and the other supported sources, reported together. Run aicharts stats --list-clients for the full list.
Start with a local report
The collector writes a report file with token counts, known costs, and source coverage. The detailed reports page reads that file in your browser tab and does not upload it.
Installing xcb or Gobstopper adds aicharts and turns on its daily history and daily update check on your computer, so aicharts history report and aicharts mcp work right away. Nothing is uploaded.
aicharts stats --home "$HOME" --all --json > usage-report.jsonWhat the numbers mean
A total counts only what each client records in its local files, and it lists the sources it covers. It isn’t a bill or a productivity score, and a counted request may have come from an agent rather than a person.
- Source coverage shown
- No transcript uploads
- Public ranking only if you opt in
