Surveillance that runs on its own isolated computer.
Feeds, detection, recording, and an autonomous watch — all inside one hardware-isolated VM. Detection happens on the VM's own CPU; no frame is ever sent to a cloud model.
The whole pipeline lives in the VM.
Most camera-AI products send your footage to someone else's GPUs. The Arkivist Computer doesn't. Ingest, detection, recording, and the autonomous swarm all run inside the same isolated machine, under your encryption keys.
That means the most sensitive data you have — what your cameras see — is governed by the same hardware boundary as everything else on the computer. No third-party vision API. No frame leaves the VM.
The pipeline - Ingest, detect, record, watch.
- Feed ingest. Camera streams land in the VM through MediaMTX. RTSP in, normalized streams out — all inside the isolation boundary, never relayed through a shared cloud.
- On-VM detection. An RT-DETR object-detection model runs on the VM’s own CPU. Frames are scored locally and detection events are written to your local database. No frame leaves the VM for inference.
- Recording & playback. Continuous fMP4 recording with retention windows and a scrubbable playback timeline — stored on your encrypted disk, served back through your tenant subdomain.
- Surveillance swarm. Lens and director loops watch the feeds autonomously, decide what matters, and email the configured owner when something does — bounded by hard caps so it never runs away.
THE MACHINE
Drone-Security tier
A near-solo VM sized for vision: 6 vCPU, 24 GB RAM, 200 GB disk — enough for MediaMTX, RT-DETR detection on CPU, and Nemotron for on-VM data ops, with isolation identical to every other Arkivist Computer.
See the watch run in your environment.
We'll provision a live Drone-Security computer and point it at a feed. You watch detection and recording happen entirely on-VM.