Privacy

Counts, not identities.

Privacy is not a feature toggle in OperaLens — it is the architecture. The system produces numeric operational signals and is designed so that identifying an individual is not part of how it works.

What the system needs

  • Presence — something is here
  • Count — how many
  • Class — person, vehicle, configured object
  • Zone — where
  • Timestamp — when
  • Event — what rule fired

What it does not need

  • Name
  • Face identity
  • Biometric profile
  • Employee or customer identity

How this works technically

The architecture behind the claims.

No facial recognition

Detection models identify object classes — “person”, “forklift” — not faces, gait, or any biometric template. There is no re-identification capability.

Ephemeral tracking only

Tracker IDs are small in-memory integers used to avoid double counting. They are recycled continuously, reset on stream restarts, and never associated with a person.

No continuous video upload

Frames are processed in memory on an edge device at the location. What is stored is numeric: metric, camera, value, timestamp. Evidence clips are retained only where an operator configures them.

On-premise processing

Video is analyzed on hardware you control, inside your network. Raw video does not need to leave the site for the listed use cases.

Minimal alert payloads

An outbound alert contains the rule name, metric, numeric value, and timestamp — nothing else. You choose which destinations exist at all.

Retention you control

Stored data is numeric time series in storage you own. Retention, authorized access, and deletion procedures are configured for each deployment.

Bring your privacy team to the first call.

The architecture answers their questions better than a brochure can.