Getting started
AI NetShell is currently in private beta. This walks through what the setup looks like once you're in — from install to your first governed change.
1. Join the beta
Sign up on the beta page. Beta waves get a direct install link and a time-limited Team-tier license so you can try the governed-write path, not just read-only AI.
2. Install and add a device
AI NetShell is a single Windows binary (Windows 10 22H2+ or Windows 11) — no gateway, no
agent fleet, nothing else to stand up. Add a device by hand, or bulk-import from CSV,
ssh_config, or a SecureCRT export. Credentials are stored in the Windows Credential
Manager, never in a config file.
3. Connect and ask the AI
Open a session over SSH, serial, or RDP and start asking questions. The AI reads config and state and answers — free, instant, no approval step, because reading changes nothing. Bring your own Anthropic or OpenAI API key in Settings; there's no AI NetShell-hosted inference.
4. (Optional) Turn on governed write for one vendor
Writes are off by default in every shipped policy. Enabling a vendor is a deliberate, scoped
opt-in — set writes: enabled and add explicit write_allowed rules for
the specific commands you want the AI to be able to propose. Every proposal still needs
to clear the dangerous-verb floor and a human's diff approval before anything reaches the device.
See Security for the full model and which vendors have write proven live
today.
5. Approve, apply, and — if you need it — roll back
Review the rendered before→after diff and the exact commands panel, then approve. A device-side snapshot was already taken before anything changed, so a bad outcome is one click from being undone.
Questions about specific behavior — license lapses, offline use, supported platforms — are answered on the FAQ.