Skip to content
BENIBIX Learn

Hermes Command Approval vs Terminal Backend

People think command approval keeps Hermes in a sandbox. It does not. Approval checks a risky command before it runs and may ask you first. The terminal backend decides where every command runs: your own machine, a container or a server far away. Approval can let a bad command through. A sandbox keeps the harm away from your files.

AI🪽 Hermes Agent

Hermes Command Approval

The check Hermes runs before a risky command. You can say yes or no.

Command approval is the check Hermes runs before a risky shell command. It compares each command with a list of risky patterns, such as deleting a folder. In smart mode, the default, a helper model lets safe ones run and stops bad ones. Unsure cases come to you. You can allow once, for the session, always, or say no. Manual mode always asks, and off mode never does.

Example
Hermes wants to run rm -rf on an old folder. A prompt shows the command, and Jack picks once so it runs one time only.
Why it matters
An agent with a shell can wipe your files in one line. This is the guard that stops it.
Full entry
AI🪽 Hermes Agent

Hermes Terminal Backend

The place where Hermes runs the commands it types.

The terminal backend is where Hermes Agent runs the shell commands it types. The default is local, on your own machine. You can also pick docker for a container, or ssh for a server far away. Cloud sandboxes such as modal and daytona work too, and plugins add more. You set it in config.yaml.

Example
Leo lets Hermes run code from strangers' bug reports. He sets the backend to docker, so nothing it runs can touch his files.
Why it matters
An agent that runs commands on your own machine can break it. A sandbox keeps the damage inside.
Full entry

In one line

Hermes Command Approval
The check Hermes runs before a risky command. You can say yes or no.
Hermes Terminal Backend
The place where Hermes runs the commands it types.