Dedicated user, read-only credentials, scoped access. Deploy h-cli the way you'd deploy any monitoring tool.
"Trust is earned, not granted."
On every server, router, or switch that h-cli will connect to, create a dedicated unprivileged user. This is not for the h-cli container — it's for the remote devices h-cli manages.
Create read-only users on your network devices. h-cli doesn't need write access to provide massive value.
Create scoped, read-only tokens for your tools. h-cli can query without modifying anything.
On the server where h-cli's containers run, restrict outbound access. Only allow connections to the systems it needs.
h-cli is the AI interface, not the security boundary.
Deploy it like any monitoring tool: read-only credentials, scoped access, restricted source IPs. The Asimov Firewall adds safety — your network controls enforce it.
Store your credentials in an environment file with restricted permissions. Never commit this to git.
Built-in credential scrambler
h-cli includes a credential scrambler that replaces all sensitive values with variable names before they reach the LLM. The AI never sees your actual passwords, tokens, or keys — it only works with references like $NETBOX_TOKEN and $HSSH_KEY. The real values are injected at execution time, outside the LLM context.
Pro tip: disable the Asimov Gate in read-only mode
If your device credentials are read-only, the remote user is locked down with rbash, and your API tokens have no write permissions — the Asimov Firewall's AI gate is an extra safety layer that adds latency on every command.
In this scenario, you can safely disable the gate for faster operations. The deterministic pattern denylist still runs (zero latency), and the remote devices physically cannot be modified regardless of what commands are sent.
Only do this if you are 100% sure all credentials and remote users are read-only. If any write access exists, keep the gate enabled.
h-cli is free and open source. Set up in minutes, start querying your infrastructure immediately.