GrayPass Tandem

Your brainprint learns every device.

Build a laptop brainprint, scan once, then watch your phone prove the same invariant core while it learns neuromotor and focus natively.

Passkeys verify the device. Magic links verify the email. MFA stacks one extra step on every login. GrayPass verifies the person. Tandem carries over only the device-invariant core the server confirms. Phone neuromotor and focus learn natively; touch and motion remain bounded sensor evidence and are never inferred from laptop keystrokes.

demo user · loading...

1 Build brainprint on this laptop live

Just interact normally - type, move your mouse, scroll, click. Your behavior streams to GrayPass in real time. The button unlocks from server-confirmed frames and signal coverage. Time is guidance, not proof.

elapsed 0s About 15s is typical; evidence decides readiness
live trust - warming up · 0 frames

The trust score will sit near zero until you build the brainprint - there is nothing to score you against yet. Open the Telemetry pill in the top-right to watch every signal being captured.

server frames 0 / 5 minimum signal coverage waiting

Waiting for the first server telemetry snapshot.

Why behavioral identity beats device identity

Passkeys are a property of the device. Lose the device, lose the key. Get a new laptop, set up a new passkey. Magic links verify whoever controls the email inbox - exactly what a SIM-swap attacker is targeting.

GrayPass verifies the human behind the keystrokes. Six modalities scored in parallel: keystroke dynamics (dwell + flight time), pointer kinematics (speed, curvature, overshoot), scroll cadence, and focus patterns. Phone touch and motion feed separate bounded touch and neuromotor paths.

The desktop-shaped signals make a 54-dimensional summary vector used by the device-invariant core. Phone-native neuromotor and touch evidence are scored separately. The summaries are non-invertible: we cannot reconstruct what you typed from the dwell-time distribution.

Templates are cancelable. A leak rotates the key, not your identity. And the device-invariant core can be compared across hardware. Tandem still learns device-native behavior separately: laptop keystrokes stay here, while phone neuromotor and focus learn on the phone. Touch remains bounded corroboration rather than a claimed learned identity template.

2 Transfer to phone

Scan the QR with your phone's camera. Token expires in 90s.

Complete Step 1 first.

3 GrayPass network waiting

Every beat below is backed by an authenticated snapshot or API response. Missing fields stay visibly waiting; laptop keystrokes are never presented as translated phone touch.

Complete Step 1 first.
Tandem event channel Connecting live updates
  1. Phone connectedWaiting for a phone session.
  2. Resonance samplingWaiting for server frames.
  3. Invariant core matchWaiting for the server gate.
  4. Temporary trust loanWaiting for loan detail.
  5. Phone-native learningTouch readiness has not been reported.
  6. Brainprint syncedWaiting for a versioned sync snapshot.
sync version Sync version waiting
freshness Sync time waiting
temporary trust loan Waiting

Source and target waiting

Baseline waiting Waiting for an authenticated snapshot

Transferability map

Core signals carry only when the response says so. Touch is learned natively.

Cross-device coreWaiting for resonance
Phone neuromotorStarts on phone
Touch + motionStarts on phone
Keystroke → touchNot translated
- network trust
Waiting for network snapshot

Waiting for phone...


phone trust (live)

- waiting

Simulate an attacker on the phone

We inject bot-shaped feature frames into the phone session and the real engine scores them. The temporary loan stays visible until a real network alert withdraws it. Only a server-critical verdict can trigger the cascade.