Agent Harness Wiki

2 min read

Hermes Agent Summary

Hermes Agent is a self-improving AI agent developed by Nous Research. It is designed with a built-in learning loop that allows it to create and improve skills from experience, maintain persistent knowledge across sessions, and operate across multiple platforms (CLI, Telegram, Discord, etc.) from a single gateway.

Quick Start & Installation

The installer supports Linux, macOS, and WSL2. It automatically handles Python, Node.js, and all necessary dependencies.

One-Line Install

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

Post-Installation

source ~/.bashrc
hermes

Key Features & Capabilities

  • Closed Learning Loop: Curates its own memory, creates autonomous skills after complex tasks, and uses Honcho for dialectic user modeling.
  • Multi-Platform Gateway: A single process supports Telegram, Discord, Slack, WhatsApp, Signal, and Email with cross-platform conversation continuity.
  • Flexible Model Support: Compatible with Nous Portal, OpenRouter (200+ models), OpenAI, Anthropic, Gemini, and custom endpoints.
  • Advanced Terminal Backends: Runs in Local, Docker, SSH, Daytona, Singularity, or Modal. Modal/Daytona allow the environment to hibernate when idle to save costs.
  • Scheduled Automations: Built-in cron scheduler for natural language reports, backups, and audits delivered to your preferred platform.
  • Research-Ready: Includes batch trajectory generation and RL environments (Atropos) for training future tool-calling models.

CLI & Messaging Commands

ActionCommand / Slash Command
Start Fresh/new or /reset
Change Modelhermes model (CLI) or /model [provider:model]
Configure Toolshermes tools
Check Usage/usage or /insights [--days N]
Stop WorkCtrl+C (CLI) or /stop (Messaging)
Update Agenthermes update
Diagnosticshermes doctor

Repository Structure & Technical Details

  • agent/: Core logic, memory management, and provider adapters.
  • gateway/: Platform-specific adapters and message routing.
  • tools/: Built-in tools including browser, file operations, and code execution.
  • skills/: Procedural memory and skill management.
  • cron/: Logic for scheduled natural language tasks.
  • nix/: Nix flake and NixOS module for hardened, reproducible deployments.

Recent Technical Updates (v0.7.0)

  • Pluggable Memory: Support for various memory providers (Honcho, Supermemory, etc.).
  • Camofox Browser: Added a local anti-detection browser backend to bypass bot detection without API costs.
  • Structured Reasoning: Improved handling for models that produce reasoning-only responses.
  • Profiles: Support for multiple isolated instances with independent configs and memories.

Migration from OpenClaw

Hermes provides a migration path for users moving from OpenClaw. It imports personas (SOUL.md), memories, user-created skills, and API keys.

hermes claw migrate
hermes claw migrate --dry-run

Contributing & Community

  • Language: Primarily Python.
  • Development Setup: Uses uv for dependency management and virtual environments.
  • RL Training: Optional integration with tinker-atropos submodule.
  • Links: Documentation, Discord, and MIT License are provided in the repository.

Graph View

Backlinks