Security
Shared hardware, clear rules
A test lab that is reachable remotely must be secure by default. NablaBox has security built into the product rather than bolted on.
Remote access uses an outbound NetBird (WireGuard) connection. The REST API listens on the local interface only until an administrator explicitly enables external access.
Three roles – admin, user and read-only reporter – plus per-DUT assignments: a user can only operate the DUTs they have been assigned to. Users can deposit their own SSH keys for device access.
Every API request is logged with source address, user, action and result – including requests that were rejected.
Passwords are stored as Argon2 hashes, sessions use signed tokens, and secrets such as the VPN setup key are encrypted at rest.
A/B updates with SWUpdate: an update is either applied completely or not at all, and the previous system stays available as a fallback.
NablaBox is developed with the EU Cyber Resilience Act in mind. A documented cybersecurity risk analysis is maintained alongside the product.
Safe to hand to an AI agent
Letting software operate physical hardware on its own needs firm limits. NablaBox enforces them on the device, not in the agent’s instructions – whether the agent uses the MCP server or the REST API:
- Create an account with the user role for the agent – it can never reboot the gateway, change the network or manage users.
- Assign it the DUTs it may work on – every other DUT rejects its requests.
- Review what it did in the audit log – every request, allowed or rejected, with user, action and result.
The DUTs stay on their isolated network, and each DUT’s console admits one session at a time, so an agent and an engineer never type into the same console.