Account
Security model
What protects your devices, and what we can and can't see.
On the device
- The agent and the
fhCLI are open source (Apache-2.0): github.com/Abnersouza7/fieldhand-agent. Release binaries are built from that source, and the README shows how to check a release's signature. - The agent only dials out over TLS 1.2 or newer. Nothing listens on the device.
- The agent does five things: shells, health reports, fleet commands, files, and connections to web pages on the device or its private network. Commands and file access run as the shell user, so they can never do more than a shell on that device.
- Connections to web pages are limited to the device and private addresses; it refuses the internet.
- Screen and live view run as the display's user, never as root when shells run as someone else, and the VNC server is reachable only by the agent.
- Updates must carry our release signature; see automatic updates.
Keys and tokens
- Device keys and install tokens are 256-bit random values. We store only their hashes.
- Removing a device revokes its key and cuts its live connection immediately.
- Each terminal needs a single-use ticket that expires in 30 seconds and names one person and one device.
Dashboard
- Sign-in by emailed link or code with a bot check, an optional password (12+ characters), and optional two-factor with any authenticator app, enforced by the API too.
- A strict Content Security Policy: no inline or third-party scripts except the bot check, and the dashboard can't be framed.
- Every shell session is logged with person, IP, time and bytes; file transfers and web pages opened, with person, IP, path or port and size.
- Device web pages open on their own domain, isolated from the dashboard, behind a cookie the device never receives.
What Fieldhand can see
Terminal traffic passes through our relay encrypted in transit; we don't store it. We store device metadata (hostname, OS, kernel, network interfaces, public IP), health samples, connection events and session metadata. Details are in the privacy policy.
If our servers were compromised
Today the relay is trusted. Someone who took over Fieldhand's servers could open shells, run commands and read files on connected devices, as each device's shell user. They could not replace the agent: updates need a release signature whose key never touches our servers. Removing a device or uninstalling the agent cuts access at once.
Locked mode (in design)
An opt-in mode where devices only obey your team, not us. Each person's browser holds a signing key that never leaves it; at install, a device pins your workspace's keys and from then on refuses any operation they didn't sign. Shells, file transfers, tunnels and screen sessions are encrypted from your browser to the device, so the relay carries only ciphertext. Fleet commands and schedules are signed when you create them and checked by each device when it runs them.
- A new browser needs approval from someone already on the team, or from a recovery key you keep offline.
- If every key is lost, devices have to be reinstalled. That's the price of us not being able to help.
- Our servers would still see metadata (who connected when, sizes, health) and could still refuse service.
- The dashboard's code is served by us, so locked mode also needs that code to come from a separately deployed, published release; a command-line signer is planned for teams that want no browser in the loop.
Want it sooner, or want to review the design? Write to us.
Found a vulnerability? Email support@fieldhand.dev. We aim to reply within two business days.
Something wrong or missing on this page? Tell us.