What it is.
Security teams are asked to trust a lot of software with a lot of footage. Semaphor is designed to earn that trust by asking for as little as possible and protecting what it does handle. It works from the monitoring screen instead of the cameras, processes detection locally, encrypts and authenticates every remote AI request, deliberately avoids facial recognition and retires old evidence automatically. Here is exactly how.
The five commitments.
1. No camera credentials, ever.
Because Semaphor works with your existing camera software by observing the rendered monitoring screen, it never needs passwords for cameras, recorders or the VMS. It does not log in to devices, open video streams or call vendor APIs. The attack surface your client’s IT team has to think about stays exactly where it was.
2. Local processing option.
Screen capture, pane separation, camera identity and person and vehicle detection run locally on the monitoring machine. The AI scene review can run locally too, on your own hardware, so no footage leaves the building. The capture service binds only to the local machine, and camera images reach a remote model only when a remote endpoint has been explicitly configured.
3. Verified, authenticated encryption for hosted AI.
When you choose hosted inference for AI scene reports, only selected event frames and their metadata are sent — never a continuous video stream — and the connection is locked down:
- Verified HTTPS only. TLS 1.2 minimum, with TLS 1.3 where supported. The certificate chain, expiry and hostname are always verified; there is no switch to turn verification off.
- Authentication required. Remote requests carry a provisioned credential in the request header — never in URLs, screenshots, request payloads or the evidence database.
- Optional mutual TLS. For stricter deployments, the client presents its own certificate, and private certificate authorities are supported.
- No fallbacks. Redirects are rejected, certificate failures never fall back to plain HTTP and never switch to another provider, and payload and response sizes are bounded.
If the connection fails, the preliminary alert stays on the guard’s screen and the AI review is marked unavailable. A network problem never declares a scene safe.
4. No facial recognition.
Semaphor detects that a person or vehicle is present and follows it with an anonymous, per-camera track number. It does not recognize faces or identify individuals. Where appearance notes are used, the vocabulary is limited to things like clothing color or vehicle type, and race, ethnicity, gender and age inference are excluded by design. Guards get what they need to act — where, when, what is happening — without building a biometric database.
5. Bounded retention.
In guard mode, retention is automatic. Working images expire in minutes, the rolling capture timeline stays short, incident images are kept for a bounded period within a storage budget, and closed incident records expire on schedule. Old material is removed on a continuous sweep, and when evidence expires the incident says so. Semaphor keeps what an operation needs to review and document an event, and no more. See evidence and audit trail.
Why it matters.
For security companies.
- Onboard clients without collecting their camera passwords.
- Answer client security questionnaires with specifics.
- Choose local or hosted AI per client.
For clients and residents.
- No facial recognition on people who walk past a camera.
- Footage stays local unless hosted AI is chosen.
- Evidence is kept for review, then retired.
Local versus hosted processing is a real trade-off in cost, speed and control; we compare the approaches in edge vs cloud video analytics.
Where it matters.
- Apartment and residential properties — residents expect security, not surveillance of who they are.
- Schools — cameras full of students, where facial recognition is a line many districts will not cross.
- Data centers — strict controls on what leaves the building and how.
For how this website itself handles contact details, see our privacy policy.
Frequently asked questions.
Does Semaphor send our camera footage to the cloud?
Only if you choose hosted AI review. Detection runs locally. With hosted inference, only selected event frames are sent, over verified HTTPS with authentication and optional mutual TLS. The AI review can also run fully on your own hardware.
Does Semaphor use facial recognition?
No. It detects people and vehicles and assigns anonymous, per-camera track numbers. It does not identify individuals.
Do you need access to our cameras or VMS?
No. Semaphor observes the monitoring screen and never needs camera, recorder or VMS credentials.
Bring your security questions to a live demo, or scope data handling for your sites in a pilot.