If the backend of the Hermes Agent SSH terminal cannot be connected, don't doubt the model first, and check the normal SSH: whether the host can be connected, whether the key is available, whether the remote working directory exists, and whether the remote machine has any dependencies required for the task. Hermes just puts the terminal execution to the remote and still relies on the SSH environment underneath.
Reproduce with the ssh command first
Perform normal ssh user@host on the terminal with the same user. If all else fails, Hermes will not be able to connect either. Common causes include overly wide key permissions, known_hosts changes, firewalls, incorrect ports, and remote shell initialization errors.
Check the Hermes configuration again
- Verify that
terminal.backendis set to ssh. - Verify that the host, user, port, and key paths are written correctly.
- The remote work directory must exist, and the permissions must allow read and write.
- If you need dependencies such as Node, Python, git, rg, etc., you can also install it remotely.
The SSH backend is good for isolating dangerous execution to remote machines, but don't hand over the production server directly to the agent. It's best to prepare dedicated users, private directories, least privileges, and test environments.
