Install and deploy

AARCV ships as one container image. The same image runs the demo on a laptop and the hardened install in a disconnected enclave.

Demo on a laptop#

Needs Docker and about 4 CPU cores. The first start renders and processes seven synthetic sorties, which takes 15 to 40 minutes depending on the machine.

git clone https://github.com/kevin-owens-product/computer-vision.git
cd computer-vision
docker compose up --build demo
# then open http://localhost:8000  (admin / demo)

Data lives in the aarcv-data volume. To update, pull and rebuild; do not remove the volume or the sorties are processed again.

Air-gapped enclave#

Built on a connected host, carried in on approved media, verified and started inside:

1. On the build host, deploy/airgap/build-bundle.sh builds the hardened app image and a TLS proxy image and writes aarcv-airgap-<version>.tar.gz with SHA-256 sums (and a GPG signature if configured).
2. Inside, install.sh verifies the sums, loads the images, checks image IDs, generates the environment file and starts the services.
3. The proxy is the only published port (TLS 1.2/1.3 on 8443). The app sits on an internal-only network with a read-only root filesystem, running as a non-root user.

Nothing in the enclave needs network access: no package downloads, no CDN assets, no license checks and no telemetry.

Sizing#

One 4-core box processes a 90-second engagement in about 10 minutes. The web app works through uploads one at a time per box; batch processing from the command line runs several in parallel. Use Fleet scale → Sites & capacity to plan boxes per base from expected sorties per day.

After install#

• Change every demo password, or set users with AARCV_USERS.
• Load your classification guide on the Security page.
• Run aarcv posture to produce the security posture evidence for your RMF package, and aarcv sbom for the software bill of materials.

Full detail is in docs/DEPLOYMENT.md in the repository; configuration options are in the developer docs.