Security by design

Vectron holds the keys to your network. Here is exactly how it treats them. This page covers Vectron for Mac, iPhone and iPad, coming to the App Store in October 2026. Last updated: 14 October 2026.

The vault

Hosts, folders, credentials, templates and IP records are stored in one vault file, in Vectron’s folder in your iCloud Drive (so the same vault opens on your Mac, iPhone and iPad) or only on that device if iCloud Drive is off. It is encrypted on your device before it is saved with AES-256-GCM, with a new random nonce every time it is saved. The key is derived from your master password with Argon2id (64 MiB of memory, 3 passes; older vaults switch to these settings on their next save) and a random salt; the settings are stored in the file, so they can be raised in future versions without breaking older vaults.

Your master password is never written anywhere unless you turn on Remember on this Mac. Then it is kept in the macOS Keychain, protected by Touch ID or your Mac password each time it is read, and never synced to iCloud or restored to another Mac. On iPhone and iPad the same applies with Face ID: the remembered password stays in that device’s Keychain.

There is no recovery. Nobody, including us, can open a vault without its master password, so choose a strong one. On the Mac, Vectron keeps the last 10 copies of your vault as backups. To keep the vault only on this Mac, turn off iCloud Drive for Vectron in System Settings › Apple Account › iCloud.

Connections

• SSH and SFTP: the first time you connect to a server, Vectron shows its host key fingerprint and asks before any password is sent (connecting to many new devices at once shows them in one list). Once trusted, the key is remembered; if it changes later, Vectron stops and warns you. Prefer not to be asked for new servers? Settings → Security → New hosts. A changed key is always shown.
• RDP: the server certificate is handled the same way: shown and confirmed on first use, remembered, and a change is shown before you sign in.
• VNC: with RealVNC RA2 or VeNCrypt (TLS) the session is encrypted and the server’s key or certificate is checked like an SSH host key. If a computer that proved its identity before stops doing so, Vectron warns you before signing in. A password is never sent unencrypted without your explicit yes.
• Telnet and some VNC servers do not encrypt traffic. Vectron supports them because network gear still needs them, and marks such sessions; use them on networks you trust.
• Broadcast: what you type at a user name or password prompt goes only to the panes that are signing in too, never to a pane at its command prompt.
• TFTP server: runs in its own small process inside Vectron’s sandbox, so a fault reached from the network cannot touch your open vault. It answers only on the network you choose (by default the wired one), or over a VPN only to devices that come through the tunnel, refuses uploads unless you allow them, stays inside the folder you picked (your home folder and Vectron’s own folders cannot be shared), never overwrites an existing file, limits uploads to 512 MB each and 4 GB per run, deletes an upload that does not finish, and logs every request, including refused ones.
• Nothing passes through servers of ours. We don’t run any.

The AI assistant

The assistant is optional, works in terminal sessions (SSH, Telnet and serial console; not remote desktops) and uses your own provider account and key. Before the first question to a provider, Vectron shows what will be sent and asks for your permission; you can withdraw it any time.

• Only read-only commands, and by default only after your click: on its own, the assistant may run only commands that read, and with Ask first on (the default) even those wait for your click. With Ask first off, it may run read-only commands (show, display, get and a checked list of server commands); every part of a piped command must be a known read-only filter, and commands with hidden characters are refused. Anything that could change a device is never run by the assistant; it is proposed to you instead.
• Read-only, checked by Vectron: the assistant may run only commands that view (show, display, get, ping and a checked list of Linux and Unix viewing commands with their harmless options). Each command is checked as text before it is typed: anything that writes, deletes, runs other programs, follows output forever, uses sudo, shell redirections or command substitution is blocked and proposed to you instead. By default the assistant also asks before each command it runs.
• Secret files are never read: on servers, files that may hold passwords or keys (/etc/shadow, sudoers, anything under .ssh, .env files, *.pem and *.key, shell histories, cloud and database credential files, /proc/*/environ and similar) are refused, also through wildcards and .. paths, on top of the output redaction below.
• You send changes: suggested commands are shown in full. Review & send types them into the session only after you confirm, and the assistant never types the same command twice for one question.
• Secrets are removed first: before anything leaves your device, Vectron replaces the secrets it recognizes (passwords, enable secrets, SNMP communities, pre-shared and RADIUS/TACACS keys, private keys, key=value settings and password hashes in common device and server formats, including Fortinet ENC values) with ‹hidden›. You can also hide IP and MAC addresses. Pattern-based removal cannot know every format, so check unusual output before you share it.
• Device output is data: the assistant is told to treat everything from the device as data, never as instructions.
• Web search (optional, with Claude, Gemini or OpenAI): only the search words go to your provider, never IP or MAC addresses. If the words came from your device (a host name, a serial number), Vectron asks first.
• On topic: a question that is not about network or system work is sent without anything from your session, and the assistant does not answer it.
• Token count and limit: Vectron adds up the tokens the provider reports and keeps the count in Vectron’s settings on this Mac; it is never sent anywhere. The optional monthly limit is a local safety net that pauses the assistant, not billing: your provider’s account is the only authority on usage and cost.
• A model on your own Mac (Ollama, LM Studio or any OpenAI-compatible server on your network, under Local server) needs no key, and nothing leaves your network.
• Other (OpenAI-compatible) services on the internet must use https, are confirmed once in a macOS dialog, count toward your monthly token limit and ask for your permission before the first request, like any other provider. Their key is stored separately and only ever goes to the address you confirmed.

On iPhone and iPad

Vectron for iPhone and iPad opens the same vault with the same encryption, unlocked with Face ID. SSH host keys and RDP certificates are checked and remembered the same way, and a changed key is shown before you sign in. Production hosts ask before connecting. The optional AI assistant works in SSH and Telnet sessions with your own provider and key: Vectron asks before anything is sent, removes the passwords it recognizes, and commands that change a device always wait for your approval. There is no account and no telemetry.

Clipboard and paste

• Reading the clipboard is only allowed while a session is open and a Vectron window is in front: for remote desktops when their clipboard sharing is on, and for terminals only when you paste yourself (⌘V, Edit → Paste, the right-click menu or right-click paste). Vectron does not read the clipboard at any other time.
• Pasting more than one line into a terminal asks first: “Paste 12 lines into DC-01?”, or “into 4 panes (…)” when Broadcast is on, with the lines shown and Cancel as the default button. This applies to every way of pasting, including dropped text; it can be switched off in Settings → Terminal.
• Copied passwords are cleared from the clipboard after 30 seconds unless you copied something else in the meantime.

What we don’t collect

Vectron’s privacy label on the App Store is Data Not Collected. Vectron has no analytics, telemetry or crash reporting service of its own. It does not report which devices you use or what you type. If you agreed on your device to share analytics with app developers, Apple may send us anonymous crash reports; that is Apple’s service, not ours. If you contact support from the app, you can attach a support report; it is a plain text file you preview first, and it never contains passwords, keys, vault contents or terminal output. See the Privacy Policy.

Permissions Vectron for Mac asks macOS for

Network access (your connections), a local network server (only for the TFTP server, and only when you start it), serial ports (console cables), the files and folders you pick and Vectron’s own iCloud Drive folder (for the vault, if iCloud Drive is on). Touch ID is Apple’s own system; your fingerprint never leaves the Mac. Nothing else.

Permissions on iPhone and iPad

• Local Network: iOS asks you once. Vectron needs it to connect to devices on your network and for the IP scanner.
• Face ID: unlocks the vault. Face ID is Apple’s own system; your face data never leaves the device.
• iCloud Drive: Vectron’s own folder, for the encrypted vault file. We cannot read it.
• Network neighbours (iOS 27 and later): lets the IP scanner show the MAC address and maker of devices it finds, by reading only the iPhone’s or iPad’s own network table while a scan runs. Nothing is sent anywhere. On iOS 18 to 26 this permission does not exist, and the scanner shows no MAC addresses.

Reporting a security issue

Email support@vectronlabs.app with the subject “Security issue”. We reply within 48 hours and keep you updated until it is fixed. Please give us a reasonable time to ship a fix before sharing details publicly. Our contact details are also in security.txt.