Hermes Agent

Hermes Agent by Nous Research is a self-improving AI agent with a built-in learning loop. It creates skills from experience, improves them during use, and builds a deepening model of the user across sessions.

Installation

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

See the Hermes quickstart for full details.

Base Invocation

Hermes is invoked in its scripted one-shot mode (-z), designed for programmatic callers:

hermes -z "<prompt>"

This produces single prompt in, final response text out — no banner, no spinner, no tool previews.

Common Flags

Flag	Type	Description
--provider	string	Override provider (e.g. openrouter, anthropic)
--model	string	Override model
--yolo	boolean	Bypass dangerous-command approval prompts
--quiet	boolean	Suppress banner/spinner/tool previews
--toolsets	string	Comma-separated toolsets (e.g. web,terminal,skills)
--skills	string	Preload skills
--checkpoints	boolean	Enable filesystem checkpoints
--max-turns	number	Maximum iterations per turn

YOLO Mode

Hermes supports a --yolo flag that bypasses dangerous-command approval prompts. This is essential for fully automated workflows where no human is present to approve commands:

steps:
  - name: auto-refactor
    action: harness.run
    with:
      provider: hermes
      prompt: "Refactor the auth module to use interfaces"
      yolo: true
      toolsets: "terminal,skills"

Generated invocation:

hermes -z "Refactor the auth module to use interfaces" --toolsets terminal,skills --yolo

Warning: This bypasses safety checks for dangerous commands. Only use in isolated or trusted environments.

Example

steps:
  - name: refactor-auth
    action: harness.run
    with:
      provider: hermes
      prompt: "Refactor the auth module to use interfaces"
      yolo: true
      toolsets: "terminal,skills"

Generated invocation:

hermes -z "Refactor the auth module to use interfaces" --toolsets terminal,skills --yolo

Fallback Example

harness:
  provider: hermes
  yolo: true
  fallback:
    - provider: claude
      model: sonnet

steps:
  - name: analyze
    action: harness.run
    with:
      prompt: "Analyze the codebase for security issues"

Notes

• Hermes supports many providers via --provider: openrouter, anthropic, openai-codex, copilot, gemini, deepseek, nvidia, and many more
• Use --quiet or --yolo for CI/automated workflows where interactive prompts are not possible

See Also

• CLI Commands Reference — All hermes subcommands and flags
• Configuration Guide — config.yaml, providers, models
• Quickstart — Install, setup, first conversation
• Toolsets — Built-in tools and toolset system
• Skills System — Creating and managing skills