Skip to main content

Security

What stays on your network​

Camera frames and inference results never leave your private network. The gateway processes images locally and returns results to the camera over your LAN. Anava's cloud services see policy checks, heartbeat, and billing signals only.

What the connection file protects​

~/.config/anava-local/camera-connection.json is a bearer credential. It contains:

  • The gateway's HTTPS endpoint
  • The fixed anava-local model alias
  • A shared secret
  • A private CA certificate

Anyone who has this file can call your gateway. Treat it like a password:

  • Do not email it, message it, or commit it to a repository.
  • Do not upload it anywhere. The installer does not upload it, and neither should you.
  • Rotate it if you suspect it was exposed. See Operate.

HTTPS-only requirement​

Camera administrator access must be HTTPS-only before you import the connection file. This is enforced, not optional: import is refused otherwise. It exists so the shared secret and certificate never cross the network unencrypted, even briefly.

No root required​

The installer never needs root access. It installs to $HOME/.local/bin and $HOME/.config/anava-local, and installs the gateway as a user-level service (launchd or systemd), not a system service.

Download verification​

The installer downloads the release archive from https://storage.googleapis.com/anava-acaps/local-gateway/vVERSION/ and verifies it against SHA256SUMS before installing anything. It rejects user-controlled symlinks in its configuration hierarchy.

Entitlement checks happen per request​

Local inference is authorized against your verified Marketplace entitlement at request time, not against a saved local setting. A saved connection file or configuration is not itself an entitlement. If your paid license lapses, is suspended, or is revoked, Local inference stops even though the gateway and camera configuration are unchanged.

No cloud fallback​

If the gateway is unreachable or the model call fails, the camera reports a Local failure. It does not automatically switch to Cloud inference. Switching to Cloud is always an explicit action you take in camera settings.

No token or usage reporting for Local calls​

Local inference requests do not report cloud inference token usage and do not debit a cloud inference budget. Your Hosted EDGE device-hour billing is unaffected by this distinction; see Overview.

Related: anava.ai/edge/local