N6C · Applied AI in the field
Project in development

Kestrel

Real-time perception for cameras and drones. Kestrel detects, identifies, and tracks objects and people on a live feed, and runs on a totally independent machine — fully local and air-gapped by default, so nothing leaves the device. Connectivity is a choice, never a requirement.

Fully localOfflineAir-gappedEdge-ready
In brief

A camera should be able to understand what it sees, and act on it, without handing a single frame to someone else's computer.

Kestrel is that understanding, built to run at the edge: it reads a live feed, finds and follows what matters, and turns it into a decision in the moment it counts — on a drone, a fixed camera, or a laptop, the same core behind each.

Because it runs on its own hardware, it holds up where connected systems don't: remote sites, sensitive facilities, and anywhere the link is thin or simply not wanted. Where a deployment calls for more reach, it scales up to meet it.

What it does

One perception core, from a webcam to a drone.

01

Detect

Real-time object detection on a live feed — bounding boxes, classes, and confidence, at interactive frame rates.

02

Track

Stable identities across frames, so a target persists through motion and brief occlusion instead of flickering.

03

Recognize

Opt-in, consent-based face recognition against an enrolled roster — for access and search, never mass surveillance.

04

Act

Autonomous drone behaviors on top of what it sees — orbit, follow, or center on a tracked target.