Hermes Agent

The Universal AI Agent for Terminal, Chat, and Beyond

NousResearch Open Source Multi-Model Multi-Platform

What is Hermes Agent?

👤 Hermes Agent Core Terminal Browser Web / MCP + 15 more

An AI agent that actually does things

  • Runs shell commands, writes and edits files
  • Browses the web, automates browsers
  • Searches codebases, manages Git & GitHub
  • Persistent memory across sessions
  • Spawns sub-agents for parallel work

Where it runs

  • CLI — interactive terminal sessions
  • Telegram / Discord / Slack / WhatsApp
  • Signal / Matrix / Mattermost
  • Cron jobs — autonomous scheduled tasks
  • Webhooks — event-driven activation

Architecture Overview

PLATFORMS CLI Telegram Discord Slack WhatsApp +4 more Gateway (Message Router) Agent Core Reasoning · Tool Loop · Memory · Skills LLM Providers (20+) Toolsets (15+) Skills (92+) Key Features ✦ Persistent memory ✦ Session search ✦ Sub-agent spawning ✦ Cron scheduling ✦ Voice (STT/TTS) ✦ MCP integration ✦ Multi-model routing

Built by NousResearch · Open source · MIT License

The Skill Document System

YAML Steps SKILL.md

Procedural memory, not just prompts

  • 92 skills across 20+ categories
  • YAML frontmatter + Markdown instructions
  • Auto-loaded when matching a task pattern
  • Self-healing — patches itself when outdated
  • Linked files: templates, scripts, references

Categories include:

MLOps GitHub DevOps Research Creative Gaming Productivity Media Smart Home Social Media Email Note-taking

Skills are the agent's institutional knowledge — they compound over time

Toolset Ecosystem

>_

Terminal

File I/O

Web Search

Browser

Vision AI

Sub-agents

Execute

  • Shell commands & scripts
  • Python code execution
  • Background processes
  • Cron job scheduling

Research

  • DuckDuckGo / Parallel search
  • Web extraction & scraping
  • ArXiv paper retrieval
  • Browser automation

Create

  • Image generation
  • Text-to-speech output
  • File writing & patching
  • Presentation & doc creation

Universal Model Support

Hermes Core Anthropic OpenAI Google DeepSeek OpenRouter Local / Ollama

20+ providers, 200+ models

  • anthropic/claude-opus-4.6
  • openai/gpt-4o / o3
  • google/gemini-2.5-pro
  • deepseek/deepseek-r1
  • meta-llama/llama-4-maverick
  • Any OpenAI-compatible endpoint

Switch models mid-session

  • CLI: hermes chat -m MODEL
  • Chat: /model MODEL
  • Per-cron-job model overrides
  • Mixture-of-Agents for hard problems
  • Automatic provider routing via OpenRouter
  • Reasoning effort control: minimal → xhigh

Deployment Options

>_

Local CLI

  • One-line install via curl
  • Interactive terminal UI
  • Session persistence in SQLite
  • Git worktree isolation mode
  • Preload skills with -s

Gateway Server

  • hermes gateway run
  • Multi-platform message routing
  • Telegram, Discord, Slack, etc.
  • Voice message transcription
  • Systemd service install

Self-Hosted VM

  • Ubuntu / Debian / macOS
  • Docker-ready (compose file)
  • Nginx reverse proxy
  • SSH remote execution
  • Webhook endpoints

Install in 60 seconds: curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

Hermes Agent vs. OpenClaw

Dimension Hermes Agent OpenClaw
Model lock-in Any LLM — 20+ providers Primarily Claude-focused
Skill system 92+ self-healing SKILL.md docs Hard-coded capabilities
Memory Persistent + session search + memory notes Context-window only
Platforms CLI + 8 messaging platforms CLI only
Scheduling Built-in cron + webhook triggers Manual invocation
Voice I/O STT (Whisper) + TTS (ElevenLabs, etc.) Text only
Multi-agent delegate_task + hermes spawning + MoA Single agent
MCP support Native stdio + HTTP client Limited / third-party
License MIT — fully open Varies

Real-World Workflows

🔬 ML Research

  • Search ArXiv for GRPO papers
  • Download & summarize findings
  • Set up training with Axolotl
  • Monitor via W&B integration
  • Push model to HuggingFace Hub

🛠 DevOps Automation

  • Clone repo, inspect codebase
  • Create feature branch
  • Implement changes + tests
  • Open PR, monitor CI
  • Auto-fix failures, merge

📊 Daily Briefing (Cron)

  • Cron triggers at 9am daily
  • Scrape news & market data
  • Summarize in Telegram DM
  • Check GitHub notifications
  • Report with voice message
92Skills
15+Toolsets
20+Providers
9Platforms
∞Possibilities

Why Hermes is the Future of AI Agents

Single LLM + Tools + Memory + Skills + Agents

The compounding agent

  • Skills accumulate — every solved problem becomes reusable knowledge
  • Memory persists — learns your preferences, environment, and conventions
  • Model-agnostic — rides every wave of LLM improvement automatically

What's coming next

  • Agent-to-agent protocols — federated multi-agent collaboration
  • RL training environments — agents that improve from experience
  • Skill marketplace — community-contributed procedural knowledge
  • Autonomous long-running missions — days-long self-directed work

The best agent is the one that gets better every time you use it.

github.com/NousResearch/hermes-agent · MIT License · Star ⭐ us