Everything you need to manage your workforce

Not photo comparison — real biometric science. AdaVision ID v3 is a deep convolutional neural network (175M+ parameters) that compresses each face into a 1024-dimensional vector signature. Matching uses cosine-similarity 1:N search across the entire enrolled workforce in a single API call. The model is trained on tens of millions of facial signatures across diverse lighting, angles, and demographics — invariant to beards, glasses, hairstyles, and different cameras. Real-world deployments hit 99-100% match scores with a 95%+ gap to runner-up candidates.

Secure 4-6 digit PIN for situations where camera access is limited. The system captures a webcam snapshot during PIN entry for audit purposes — giving you visual verification even without facial matching.

Verify workers are where they should be. Set up geofence boundaries around your sites and track GPS coordinates at every clock-in. Choose between warning mode for soft enforcement or strict mode to block off-site clock-ins entirely. Three modes per location: Disabled (no GPS checks), Warning (records position but allows clock-in), and Strict (blocks clock-in if outside geofence).

Every clock-in records a precise confidence percentage. Admins see exactly how strong each verification was — typically 90%+ for genuine matches. Configurable thresholds let you set the security level per site.

See every active kiosk terminal in real time. Monitor session status, clock-in activity, and watch live camera feeds from any kiosk — whether it has a location assigned or not. Per-session feeds, session health monitoring, clock-in/out counters, kiosk lock status, and IP address tracking.

A real-time fleet dashboard for every active kiosk. RAG status (green / amber / red), AWS error counts in the last 5 minutes, 1-hour match success rate, and last clock-in time per device. Push a software update? Click Reload on a single kiosk, all kiosks for one client, or the whole fleet — every reload lands on its target within ~60 seconds via the kiosk's next health beacon. No physical visit, no remote-desktop session.

Each kiosk device has a stable identifier in localStorage. Admins can pin that identifier to a location in the Kiosk Fleet view, and from that moment on every camera frame the kiosk uploads is automatically routed to the correct site's siteboard — regardless of which URL was loaded or whether the browser cache was cleared. Eliminates the "the camera is showing the wrong site" support call for good.

Locations can have their own siteboard password. Anyone wanting to view that location's public siteboard — or its live camera feed, even via direct URL — has to enter the password once per browser. Logged-in admins and tenant managers bypass automatically. Unlock cookies are signed with a per-install server secret, marked Secure + HttpOnly, and valid for 30 days. Locations without a password stay publicly viewable.

Drop a pin on an interactive map (Leaflet + OpenStreetMap) and set a geofence radius anywhere from 20m to 10km. Warning mode logs out-of-bounds clock-ins and shows the worker an in-kiosk message; strict mode refuses the clock-in entirely. Every entry is tagged with GPS coordinates plus device accuracy — readings the system can't trust (poor signal, tablet on Wi-Fi only) get flagged 📡 on timesheets so admins can investigate. Android kiosk APK ships with location permissions baked in.

Set up a survey (yes/no, 1-5 rating, or free text) that fires after badge entry and BEFORE the face or PIN step. Mark a specific answer as blocking — for example, "No" to "Are you fit to work today?" — and the system refuses the clock-in, tells the worker to see their supervisor, and emails the alert to site admins. Every blocked attempt is recorded for audit. Non-blocking surveys still buffer the answers and store them against the resulting time entry.

Click any kiosk row in the Fleet view to slide in the Remote Tools panel: live screen preview, reload / refresh / screenshot / ping commands, and a "dump full state" action that emails the admin who clicked it a complete diagnostic (camera state, last heartbeat, JS version, kiosk_id, IP). Built without RustDesk or any VPS dependency — pure TempClock infrastructure piggybacking on the existing heartbeat. Includes a self-test page that audits the whole chain (queue, heartbeat, executor, mail) per-kiosk.

RFC 6238 TOTP via any standard authenticator app (Google Authenticator, Microsoft Authenticator, 1Password, Authy, Bitwarden). Each user gets 10 backup codes for recovery. Admins can schedule a rollout date that turns 2FA from optional to required — the system runs a notification campaign in the days before, nudges every unenrolled user, and shows enrolment stats in real time. 30-day trusted-device cookies signed with HMAC-SHA256. Backup codes are stored as bcrypt hashes; TOTP secrets are encrypted at rest. Mandatory across admin_users, tenant_users, and client_users.

The entire AdaVision ID v3 inference pipeline runs on our EU-resident infrastructure (Ireland data centre). Vector signatures and Liveness session data never leave the EU — full UK-GDPR data residency. Sub-400ms typical recognition latency. Each tenant has a physically isolated vector collection: cross-tenant matches are technically impossible, not just policy-protected.

AdaVision ID v3 uses ISO/IEC 30107-3 Presentation Attack Detection (PAD level 1 certified) — the same standard adopted by banking and identity-verification platforms. Workers complete a single fluid liveness challenge on their phone: head movement and randomised chromatic flash sequence. Photo replays, video loops, deepfakes and printed/silicone masks are rejected at the model level. Takes around 30 seconds; no manual pose-by-pose capture.

Every enrolment is scored on sharpness, brightness, pose and frontal-ness before the vector signature is committed. Substandard captures are rejected with a clear retry prompt — workers retake the check rather than living with a marginal template. This is the lever that locks in real-world match accuracy: a clean enrolment template gives you 99%+ matches for the rest of the worker's tenure.

When a worker faces the kiosk, AdaVision ID v3 doesn't ask "is this a match for badge #1234?" — it asks "who is this, across our entire enrolled workforce?" and returns the top candidate with a similarity score. This 1:N model defeats badge-swap attacks: even if someone enters another worker's badge, the face match identifies the actual person. Top candidate plus runner-up confidence are both logged for every clock-in.

Workers receive a secure, time-limited link to re-enroll their face data from their own phone. New beard, new glasses, new hairstyle — they can update their enrollment independently. Admins are notified and can review or approve changes.

Every facial recognition parameter is configurable per location via the admin config panel. Adjust detection sensitivity, match thresholds, camera resolution, enrollment requirements, and retry behaviour independently for each site. A warehouse with harsh fluorescent lighting can have different settings to a bright office.

Every clock-in event logs the precise match distance and confidence percentage. This data appears in timesheets, the live dashboard, and audit reports — giving you complete transparency on verification quality across your entire workforce.

Every clock-in and clock-out captures the device GPS coordinates. Combined with configurable geofencing — warning, strict, or disabled modes per location — you get both location verification and a complete audit trail of where every clock event occurred.