Overview

Hermes Agent is the open-source AI agent that grows with you. Built by Nous Research and released under the MIT license, Hermes Agent is an autonomous AI agent featuring a built-in learning loop — it creates skills from experience, improves them during use, persists knowledge across sessions, and builds a deepening model of who you are.

What is Hermes Agent?

Hermes Agent is a free, open-source, self-improving AI agent that connects large language models (Claude, GPT, Gemini, DeepSeek, Llama, and 200+ models via OpenRouter) to messaging platforms (Telegram, Discord, Slack, WhatsApp, Signal, Email, and more) and enables AI to take real actions through 47 built-in tools across 19 toolsets.

Key differentiator: Hermes Agent has a built-in closed learning loop:

Creates skills autonomously from experience (after 5+ tool calls, error recovery, user corrections)
Improves skills during use through continuous refinement
Persists knowledge in MEMORY.md and USER.md files across sessions
Searches past conversations via FTS5 full-text search
Models user preferences via Honcho dialectic integration

Key Features

Core Capabilities

Full TUI terminal interface: Multiline editing, slash-command autocomplete, streaming output, token/cost tracking
Multi-platform messaging gateway: 17+ platforms (Telegram, Discord, Slack, WhatsApp, Signal, Email, Home Assistant, Matrix, Mattermost, DingTalk, Feishu, Lark, WeCom, WeChat, BlueBubbles, Signal, webhooks)
47 built-in tools: Web search/extract, terminal/files, browser automation, vision/image generation, memory, session search, cron scheduling, delegation, code execution, MCP, Home Assistant, RL training
6 terminal backends: local, Docker, SSH, Daytona, Singularity, Modal — from laptop to serverless cloud
Subagent delegation: Spawn isolated child agents for parallel workstreams (up to 3 concurrent, depth limit 2)
Smart model routing: Automatic routing between primary and cheap models based on message complexity
Context compression: Automatic lossy summarization when approaching token limits
Mixture of Agents: Multi-model orchestration with 4 parallel reference models + aggregator
MCP integration: Connect any MCP server; Hermes can also serve as MCP server
Voice mode: Push-to-talk TTS/STT (6 TTS providers, 4 STT providers)
Seven-layer security model: Command approval, container isolation, SSRF protection, Tirith scanning, secret redaction, DM pairing, context injection protection

Architecture Components

CLI (hermes): Python-based terminal UI with prompt_toolkit multiline editing
Gateway: Messaging server connecting 17+ platforms from single process
Tools: 47 capabilities organized into 19 toolsets (enable/disable per context)
Memory: Persistent MEMORY.md + USER.md, FTS5 session search in SQLite
Skills: Self-improving procedural memory with progressive disclosure (3 levels)
Delegation: Isolated subagent spawning with restricted toolsets
Terminal: Pluggable backends from local shell to serverless cloud containers

Prerequisites

Required: Git (only system prerequisite — installer provisions everything else)
Required: Linux, macOS, WSL2, or Android/Termux (Windows native unsupported)
Required: Basic command-line proficiency
Helpful: Familiarity with YAML syntax and at least one LLM API
Not required: Prior Hermes Agent or OpenClaw experience

Quick Start

# One-line installation
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
 
# Reload shell
source ~/.bashrc
 
# First-time setup wizard
hermes setup
 
# Start interactive session
hermes

See the By Example tutorial for 80 heavily annotated examples covering 95% of Hermes Agent.

How This Tutorial Is Organized

By Example

Learn Hermes Agent through 80 self-contained, heavily annotated examples:

Beginner (Examples 1-27) — Installation, CLI, YAML configuration, tools, memory basics (0-40% coverage)
Intermediate (Examples 28-54) — Skills, messaging channels, delegation, scheduling, browser, code execution (40-75% coverage)
Advanced (Examples 55-80 plus four deep-dive hardening patterns) — Terminal backends, security (OWASP LLM Top 10 threat model, indirect prompt-injection defense, MCP supply-chain vetting, link-preview exfiltration prevention, network egress isolation), MCP, voice, production deployment (75-95% coverage)

Why Hermes Agent Matters

Most AI agent frameworks are reactive — they answer your current question and forget everything after the session ends. Hermes Agent is different. The closed learning loop means your agent compounds knowledge over time: today's hard-won debugging lesson becomes tomorrow's auto-applied skill, and next month's onboarding documentation.

This architectural choice has practical consequences:

Lower token costs: Skills and memory replace repeated context, cutting inference costs
Better personalization: User profiles accumulate preferences, eliminating repeated corrections
Team knowledge sharing: Skills can be shared across teams via Skills Hub
Production-grade security: Container isolation, approval modes, and secret redaction make it safe to grant real tool access

For developers switching from OpenClaw, Hermes Agent provides a built-in migration tool (hermes claw migrate) that imports your configuration, memory, skills, and messaging platform settings.

Migrating from OpenClaw

Hermes Agent includes a one-command migration path:

hermes claw migrate --dry-run           # Preview migration
hermes claw migrate --preset full       # Full migration including secrets
hermes claw migrate --preset user-data  # Exclude API keys

See Example 27 for the complete migration guide with directory mapping and configuration format conversion.

OpenClaw — Alternative AI agent platform focused on tool-chaining
Claude Code — Anthropic's AI-powered coding assistant

Next Steps

Start with the By Example tutorial to master Hermes Agent through 80 heavily annotated, runnable examples.

Overview

Command Palette