Tap through your year
A story you can screenshot and post, plus a summary card and a wide card for X.
Example numbers. Yours come from your own reports.
Play a sample with made-up numbers →Your keys stay on your computer
You shouldn't have to trust us, so here's how to check every claim.
Two hosts, read-only
Keys go only to api.openai.com and api.anthropic.com, in the header of GET requests for cost and usage reports. One file, src/http.js, does all networking and refuses any other host or redirect.
Nothing sent anywhere else
No telemetry, no analytics, no update check. Your numbers and keys are never saved. The recap page is a local file that loads nothing.
Zero dependencies
The package is only the code in the repo, so nothing else in the install can phone home. A test fails the build if that ever changes.
Published from the repo
Each release is built by GitHub Actions with npm provenance, so the npm page links to the exact commit that made it.
# see every request it would make, and send none
npx tokens-wrapped --dry-run --share
Safest of all: make a fresh admin key, run it, then delete the key. Read SECURITY.md
Getting an admin key
Cost and usage reports need an organization admin key, not a regular API key.
OpenAI
- platform.openai.com, Settings, Organization
- Admin keys, Create new admin key
- It starts with
sk-admin-
Anthropic
- console.anthropic.com, Settings
- Admin keys, Create admin key
- It starts with
sk-ant-admin
Questions
Is it free?
Yes. It's open source under the MIT license.
What does it need?
Node 18 or newer, and an OpenAI or Anthropic admin key (or both). Without keys you can type your monthly spend instead.
Which dates does it cover?
January 1 to yesterday, in UTC days, read from the providers' daily reports.
Does the Claude Code card include my Pro or Max plan?
No. Anthropic's report covers Claude Code billed to your API organization, not personal or team subscriptions.
What about Codex?
OpenAI doesn't offer an API for Codex lines or commits yet, so the "built" card is Claude Code only. OpenAI tokens and spend are in every other card.
Does it read my prompts?
No. It only reads cost and usage reports: dollars, token counts by model and day, and Claude Code activity counts. Teammates' emails in the Claude Code report are counted, never shown.
Who made it?
The team at ernd.
