v2026.3.30 · Updated Mar 30, 202618.5K+ GitHubMIT · Open source

The Hermes Agent Cheatsheet

An autonomous agent that runs on your server, remembers what it learns, and grows over time — install, CLI, and gateway essentials on one page.

📄 Print / Save PDFShare on 𝕏⭐ Star on GitHubNous Research — Hermes

Maintaining this reference takes time — thanks if you can chip in.
☕ Buy me a coffee · other ways to support

🚀 Get started in 60 seconds (from the official site)

1
Install — pulls uv, Python 3.11, clones the repo, no sudo.
2
hermes setup — interactive wizard · hermes model — pick Nous Portal (OAuth), OpenRouter (API key), or your own endpoint.
3
hermes — full interactive CLI with tools, memory, and skills.
4
Optional: hermes gateway setuphermes gatewayhermes gateway install (Telegram, Discord, Slack, WhatsApp; systemd service on Linux).
5
hermes update — pull latest changes and reinstall dependencies.
📌

What is Hermes Agent?

From the product site: it is not a coding copilot chained to an IDE or a thin wrapper around one API. It is an autonomous agent that lives on your machine or server, persists memory, generates skills from experience, and improves the longer it runs — with a real CLI, optional messaging gateway, sandboxed execution, and multi-model tooling.

Highlights

  • Multi-channel — same agent on Telegram, Discord, Slack, WhatsApp, Signal, email, and CLI via one gateway.
  • Memory & skills — remembers projects and past fixes; creates and installs skills (including community hubs).
  • Automation — natural-language cron for unattended jobs through the gateway.
  • Delegation — subagents with isolated conversations, terminals, and Python RPC for parallel work.
  • Sandboxing — multiple backends (local, Docker, SSH, Singularity, Modal) with hardening options.
  • Web & media — search, browser automation, vision, image generation, TTS, multi-model reasoning.

Platform support

Official installer targets Linux, macOS, and WSL2. Native Windows is called out as experimental — use WSL2 for a supported path. Source: hermes-agent.nousresearch.com.

Install

One-line install from the main branch script: installs tooling (including uv), Python 3.11, clones the project, and sets up the environment — typically without sudo.

One-liner

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

After install

  • Run hermes setup for the interactive wizard.
  • Run hermes to start the interactive CLI.
  • Use hermes update anytime to sync upstream and refresh dependencies.
🎛️

Configure models & auth

Model providers (from the site)

SourceNotes
Nous PortalOAuth sign-in through the Nous portal.
OpenRouterBring your OpenRouter API key.
Your endpointAny OpenAI-compatible HTTP API you operate.

Commands: hermes setup (wizard) · hermes model (choose / change model configuration).

Sign up

Account and billing flows are hosted on the Nous portal — use the Sign Up on Nous Portal call-to-action on the official landing page.

⌨️

CLI essentials

Day-to-day

hermes              # Interactive session (tools, memory, skills)
hermes setup        # Configuration wizard
hermes model        # Model / provider selection
hermes update        # Pull latest + reinstall deps

Documentation & source

For subcommands and flags beyond this summary, use upstream Features and Docs on hermes-agent.nousresearch.com and the GitHub repository.

🛰️

Messaging gateway

Connect Telegram, Discord, Slack, or WhatsApp through a single gateway process. The site documents an interactive gateway wizard and optional install as a system service (for example systemd on Linux).

Commands

hermes gateway setup    # Interactive gateway wizard
hermes gateway        # Run the gateway
hermes gateway install # Install as a system service

Why it matters

Start on CLI, then pick up the same agent on chat apps — scheduled automations and channel delivery run through this gateway layer per the official docs.

Feature themes

Lives where you do

Telegram, Discord, Slack, WhatsApp, and CLI from one gateway — switch surfaces without rebuilding the agent.

Grows the longer it runs

Persistent memory plus auto-generated skills — learns your projects and retains how problems were solved.

Scheduled automations

Natural-language cron for reports, backups, and briefings unattended via the gateway.

Delegates & parallelizes

Isolated subagents: separate conversations, terminals, Python RPC — lower context cost for pipelines.

Real sandboxing

Backends include local, Docker, SSH, Singularity, and Modal — hardened containers and namespace isolation.

Full web & browser control

Search, browser automation, vision, image generation, TTS, and multi-model reasoning in one stack.

🧰

Tools (40+ built-ins)

Marketing copy on the site lists 40+ built-in tools spanning search, terminal, filesystem, browser automation, vision, image generation, TTS, code execution, subagent delegation, memory, planning, cron, multi-model routing, and more. Treat this cheatsheet as a map — exact tool names live in repo docs.

Bundled categories (summary)

AreaExamples (from site)
Web & browserSearch, browsing, snapshots, automation
ExecutionTerminal, sandboxes, Python RPC / code run
Memory & plansLong-term recall, task planning
Omni-channelGateway integrations, schedules
MediaVision, image gen, TTS
📱

Platforms

Where you can talk to it

Telegram · Discord · Slack · WhatsApp · Signal · Email · CLI — all through the same Hermes gateway architecture described on hermes-agent.nousresearch.com.

🖥️

Run environmentsEnvironments

Where it can execute

The site highlights local installs plus remote / isolation options such as Docker, SSH, Modal, Daytona, and Singularity, with hardened containers (read-only root, dropped capabilities, namespaces). Match this list to your org’s safety requirements.

🧩

Skills ecosystem

Bundles & sharing

Hermes ships many bundled skills (site copy cites 40+ across MLOps, GitHub flows, research, and more). Agents can author new skills dynamically. The interchange format aligns with openagentskills.io, with community installs from ClawHub, LobeHub, and GitHub — as listed on the landing page.

Practice

Start from bundled skills, fork for your stack, and publish extras so every Hermes instance you run shares the same muscle memory.

🔬

Research workflows

Batch & RL hooks

Nous positions research-oriented flows: batch trajectory generation with parallel workers and checkpointing, Atropos integration for RL training, and exports to ShareGPT-style formats for fine-tuning with trajectory compression — see the Research blurb on the official site for the latest scope.