hermes setup wizard hangs on non-interactive SSH / headless environments

[sudoingX]

Bug

hermes setup wizard hangs indefinitely on non-interactive SSH sessions and headless environments (GPU nodes, Docker containers, CI/CD). The interactive provider selection menu renders but cannot receive input, blocking the entire setup flow.

Steps to Reproduce

1. SSH into a headless GPU node (no interactive terminal)
2. Install hermes-agent
3. Run hermes setup or hermes (auto-triggers wizard when no provider configured)
4. Wizard renders but hangs at provider selection — no way to proceed or skip

Select your inference provider:
    ↑/↓ Navigate  Enter Select  Esc Skip  Ctrl+C Exit
  ○ Nous Portal API key (direct API key access)
  ○ Login with Nous Portal (Nous Research subscription — OAuth)
  ...

Environment

• Ubuntu 22.04 (Docker: pytorch/pytorch:2.5.1-cuda12.4-cudnn9-devel)
• Python 3.11.10
• Hermes Agent v1.0.0
• SSH non-interactive session to GPU node

Workaround

hermes config set works perfectly in non-interactive mode:

hermes config set model.provider custom
hermes config set model.base_url http://localhost:8080/v1
hermes config set model.default your-model-name

This bypasses the wizard entirely. But users don't know this exists because the wizard blocks them before they can discover it.

Suggested Fix

Either:

1. Detect non-interactive TTY (if not sys.stdin.isatty()) and skip wizard with a message pointing to hermes config set
2. Add a --non-interactive or --headless flag to hermes setup
3. Both

Context

This blocks anyone deploying Hermes Agent on remote GPU nodes via SSH, which is a common setup for local inference workflows. Found this while setting up Hermes Agent with Qwen 3.5 9B on an RTX 3060 via llama-server.

Related: #898 (duplicate subparser crash) was blocking users before they could even reach the wizard. Now that #898 is fixed, this will be the next blocker for headless deployments.