Fortress for developers
The SDK runs on your own machine. No account, no API key, no data leaves your network. Everything below runs in this page, with the same code the SDK ships, and every number carries its date and method.
Your signing secret — instead of an API key
There is no Fortress server to call, so there is no API key to manage. What you keep is one secret that makes your marks yours. Create it here; it is generated in your browser and never sent anywhere.
Store it like a password (for example in your vault, as FORTRESS_MERCHANT_SECRET). Lose it and you can no longer prove which marks are yours. Self-service sign-up, a hosted key dashboard and metered billing are not offered: the SDK is delivered under NDA as a private repository.
Is our code in theirs?
Put your file on the left and the suspect file on the right, then press Compare code. Renaming, reformatting and deleting comments do not hide a copy. The score says how much of your file is in theirs.
Proves copy origin, not who leaked it: every licensee holds the same file, so a leak matches all of them equally (measured: 5 licensees, 5 identical scores). A full rewrite escapes the fingerprint. “No match” never means “not copied”.
Measured on real code (04.11.2026)
| Case | Found | Right file on top |
|---|---|---|
| Untouched copy · reformatted · comments removed · identifiers renamed · all three | 59/59 each | 59/59 each |
| 75 % of our file pasted into a foreign file | 59/59 | 58/59 |
| 50 % pasted | 57/59 | 51/59 |
| 25 % pasted | 18/59 — below the threshold by design | 12/59 |
| False matches, unseen third-party code (held-out) | 1 of 1,593 pairs (score 0.407, threshold 0.4) | |
origin-code-032.cjs.
Mark a whole folder
Point the CLI at a folder — local disk, a NAS mount or a synced bucket. Every image is marked at its original resolution, in parallel, and a manifest (file → key) is written for your recipient registry. Broken files are listed, never skipped silently.
fortress mark-dir ./archive ./archive-marked agency-recipient-01 --workers 8
| Workers | 24 real photos, 16 MP each | Files / s | Megapixels / s |
|---|---|---|---|
| 1 | 268 s | 0.09 | 1.08 |
| 4 | 126 s | 0.19 | 2.29 |
| 8 | 115 s | 0.21 | 2.51 |
Measured 04.11.2026 on one machine (32 logical cores, AMD Ryzen AI Max+ 395, other load present): decode, mark, lossless PNG write — the full path, not compute alone. Throughput depends on the machine and its load. All 24 marked at native resolution; worst PSNR 39.2 dB. Re-reading 6 outputs found the right key in 6 of 6; the 6 unmarked originals read no key.
Per-viewer forensics in your video player
One stored video, a different invisible mark per viewer, applied in the player. Re-measured on 04.11.2026:
| What | Result |
|---|---|
| Codecs: H.264 CRF 23/28, HEVC CRF 28, VP9 CRF 32, AV1 CRF 32 × 5 content classes | Viewer identity read in 25/25; unmarked controls 25/25 silent |
| Detection with the public anchor (1024×1792 frame) | 190 ms on one core (a blind search is estimated at ~6 hours — an estimate, not a measurement) |
| Is the live player the built one? | Yes: all 7 of 7 live player files are SHA-256 identical to the build |
Player SDK 0.95.1, desktop browser. The 4-byte payload identifies the viewer through your account registry. The React Native entry point is not measured on a phone.
The film: why, and every button explained live
First the why as a drawing, then a real recording of this page: every button is pressed and the result shown. At the end: what the product does not do. Voice: neural text-to-speech. Deutsche Fassung (mp4)