An impression, defined.
- Rule 1
One per turn, per format
A Claude turn starts when you send a prompt and ends when Claude finishes. Each turn counts at most one Prompt Banner and one Theater impression. If more than one ad of a format showed during the turn, the one that was on screen longest is the one counted.
- Rule 2
On screen long enough
A Prompt Banner counts after 1 second continuously on screen, the MRC display viewability standard. Theater, the larger slot, needs 2 seconds. The mod measures the longest unbroken stretch and doesn't report anything shorter. Our server checks that stretch against when the turn began.
- Rule 3
A turn our server issued
When a turn starts, the mod asks our server for a signed turn token, and every paid impression must carry one. A turn we never issued earns nothing and costs nothing. One device gets at most one turn every 10 seconds, and one account at most 600 turns a day.
- Rule 4
A paid ad we actually served
Each paid ad arrives with a serve token for that account and device, and an impression without a matching token isn't counted. Neither is one from a version of the mod too old to earn.
- Rule 5
Fresh
An impression that reaches us more than 24 hours after it was shown doesn't count.
- Rule 6
Inside the daily caps
Each developer is credited for at most 120 banner and 20 Theater impressions per UTC day. Impressions over a cap are counted as filtered traffic: the developer isn't credited and the advertiser isn't charged.
The mod doesn't detect window focus. It can't see your keystrokes, by design, so a turn starting is our signal that a person is there: someone sent a prompt.
Two click counts, no pixel.
- Requests from known bots and crawlers, by user agent
- Anything other than a plain GET, such as link previews that send HEAD
- Clicks less than 1 second after the ad was served, faster than a person reads
- Repeat clicks on the same campaign from the same network on the same day
The redirect sets no cookie and there's no tracking pixel. The advertiser's page gets UTM tags and nothing else: utm_source=tokenbreak, utm_medium set to terminal or desktop, the campaign and the format.
What the mod never reads.
This is written into our terms as a permanent promise, and it won't change. The mod is open source, and its test suite fails if it ever hooks the prompt, the conversation or tool calls.
Read the source →- Your prompts, or anything you type
- Claude's replies and the rest of the conversation
- Your code and any file on disk
- Tool calls: file reads, edits, shell commands
- File names, repo names, branch names
- Environment variables and secrets
- Your terminal history
- Your IP address in readable form (we keep only a keyed hash, for abuse checks)
Every request the mod sends.
- R-01
Asking for ads
About every 15 minutes while you work.
The device id is random, made on your machine when the mod was installed. X-Tokenbreak-Surface says whether the ads will be drawn in a terminal or in the desktop app's Code tab, which shows text-only ads.
requestGET /api/v1/ads?formats=banner,theater Authorization: Bearer EXAMPLE-sign-in-token X-Tokenbreak-Device: 11111111-2222-4333-8444-555555555555 X-Tokenbreak-Version: 0.6.0 X-Tokenbreak-Surface: terminal
response{ "ads": [ { "id": "00000000-0000-4000-8000-00000000ac3e", "format": "banner", "brand": "Acme DB", "glyph": "▲", "headline": "Postgres branches in 2 seconds. Free tier.", "cta": "Try it free", "accent": "#2BB673", "clickUrl": "https://tokenbreak.dev/c/EXAMPLE-serve.EXAMPLE-signature", "isHouse": false, "campaignId": "00000000-0000-4000-8000-00000000ac3e", "serveToken": "v1.EXAMPLE-serve-ids-and-expiry.EXAMPLE-signature" } ], "ttlSeconds": 900, "servedAt": "2026-10-03T21:31:12.000Z", "mod": { "latest": "0.6.0", "min": "0.6.0" } } - R-02
Starting a turn
Once, when you send a prompt.
It says that a turn began on this device, and nothing about what you asked. The answer is the turn token that this turn's impressions carry.
requestPOST /api/v1/turns Authorization: Bearer EXAMPLE-sign-in-token X-Tokenbreak-Device: 11111111-2222-4333-8444-555555555555 X-Tokenbreak-Version: 0.6.0 X-Tokenbreak-Surface: terminal Content-Type: application/json { "deviceId": "11111111-2222-4333-8444-555555555555", "surface": "terminal" }response{ "turnId": "turn_EXAMPLE_0001", "turnToken": "v1.EXAMPLE-turn-id-and-time.EXAMPLE-signature", "issuedAt": 1791063340000 } - R-03
Reporting impressions
In batches, every few minutes.
Which ad was shown, when, in which format and where, for how long, and the two tokens our server issued. That's the whole record.
requestPOST /api/v1/impressions Authorization: Bearer EXAMPLE-sign-in-token Content-Type: application/json { "deviceId": "11111111-2222-4333-8444-555555555555", "modVersion": "0.6.0", "impressions": [ { "adId": "00000000-0000-4000-8000-00000000ac3e", "format": "banner", "at": 1791063354000, "turnId": "turn_EXAMPLE_0001", "serveToken": "v1.EXAMPLE-serve-ids-and-expiry.EXAMPLE-signature", "turnToken": "v1.EXAMPLE-turn-id-and-time.EXAMPLE-signature", "surface": "terminal", "shownMs": 12400 } ] } - R-04
Reporting or hiding an ad
Only when you run /ads report, or press ✕ on an ad.
Reasons are scam, misleading, offensive, broken, other, with an optional note you write. 3 different developers reporting one campaign within 7 days pause it until a person reviews it. ✕ sends reason hide, which pauses nothing: it hides that campaign on your machine for 30 days, and the advertiser sees only a count.
requestPOST /api/v1/reports Authorization: Bearer EXAMPLE-sign-in-token Content-Type: application/json { "adId": "00000000-0000-4000-8000-00000000ac3e", "serveToken": "v1.EXAMPLE-serve-ids-and-expiry.EXAMPLE-signature", "reason": "misleading", "note": "The free tier needs a card." }✕ sendsPOST /api/v1/reports Authorization: Bearer EXAMPLE-sign-in-token Content-Type: application/json { "adId": "00000000-0000-4000-8000-00000000ac3e", "serveToken": "v1.EXAMPLE-serve-ids-and-expiry.EXAMPLE-signature", "reason": "hide" }
Signing in with /ads login and out with /ads logout talks to our sign-in service and carries only your sign-in code or token. What we keep, and for how long, is in the privacy policy.