← BACK
OFFLINE
DIFFERENTIAL PRIVACY
CONSTELLATION
Opt-in to improve models. Only noisy, differentially private patterns leave your device. Never raw text. Never PII.
Constellation Opt-In
Fully offline. No data leaves your device.
CONTRIBUTIONS
0
updates sent
PRIVACY BUDGET
ε = —
per contribution
MODEL
—
—
DP POLICY
Loading...
MODEL INFO
Loading...
HOW IT WORKS
1
Local Processing
All inference happens on your device. Raw prompts never leave.
2
Gradient Clipping
Gradients are clipped to bound sensitivity before any sharing.
3
Gaussian Noise
Calibrated noise is added to provide (ε, δ)-differential privacy.
4
Secure Aggregation
Only aggregated, noisy updates reach the server. Individual contributions are mathematically unrecoverable.
Constellation uses formal differential privacy guarantees. Read the whitepaper →