Setup and first run

The safest way to approach Hermes is in stages: install it, confirm the CLI actually works, configure the provider and model intentionally, verify session behavior, and only then expand into messaging surfaces or riskier approval modes.

1. Install Hermes on the right platform

For Linux, macOS, WSL2, or Termux, the documented installer path is:

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

For Windows PowerShell, use:

iex (irm https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.ps1)

WSL2 is the most battle-tested Windows route in the current source material.

2. Reload the shell and verify the command exists

• source ~/.bashrc
• source ~/.zshrc
• hermes

If you get hermes: command not found, treat that as a shell or PATH verification problem first.

3. Configure provider and model intentionally

• hermes model
• /model gpt-5.4 --provider openrouter --global

Hermes documents ~/.hermes/config.yaml as a real configuration surface for providers, models, API keys, and related settings.

4. Treat first success as more than “it launched once”

• the CLI launches locally
• the intended model/provider pair is visible
• session continuation works intentionally
• the approval posture is understood before broader use

5. Verify session continuity before expanding

• hermes --continue
• hermes --resume <session-id-or-title>
• hermes sessions list
• hermes sessions prune

Related pages

• Workflow and session operations
• Model configuration
• Security and approval modes
• Validation and observability