Skip to main content
🤖 Running agents for a team, not just yourself? Get an independent review of identity, secrets, failover, observability and governance. Assess your agent platform
AI

What is Hermes Agent? Nous Research's Self-Improving Agent

Hermes Agent is Nous Research's open-source AI agent that lives on a server, remembers across sessions, and improves its own skills as it works.

Luca Berton

· AI Advisor & Architect

Published · Updated · 5 min read

What is Hermes Agent?

Hermes Agent is Nous Research’s open-source, MIT-licensed take on a personal AI agent that doesn’t live on your laptop. Instead of a chat tab you open and close, it’s a process that runs continuously on a server, keeps a persistent memory of what it’s learned about you and your projects, and reaches you wherever you already are — Telegram, Discord, Slack, WhatsApp, Signal, or a plain terminal.

Nous Research describes it as the only agent with a “built-in learning loop” — it creates skills from experience, improves them with use, and builds a deepening model of who you are across sessions, rather than starting from zero every conversation.

Key Features

  • Persistent memory — a memory database under ~/.hermes/memory/ that survives restarts, so the agent remembers past conversations, projects, and preferences instead of forgetting everything between sessions.
  • A large bundled skills library — ~160 skills ship out of the box, covering GitHub workflows (PR review, issue triage), document handling (PDF, DOCX, XLSX, PowerPoint, Notion), research (arXiv, web search), coding tasks (debugging, TDD, code review), and more. The agent can also author new skills for itself as it solves novel problems.
  • Model-agnostic — 300+ models across Nous Portal, OpenAI, Anthropic, OpenRouter, and custom/self-hosted endpoints. Swap providers anytime with hermes model, no code changes.
  • Messaging gateways — 20+ documented platforms, including Telegram, Discord, Slack, WhatsApp, Signal, Google Chat, Matrix, MS Teams, LINE, Feishu/Lark, and Home Assistant, so the agent is reachable from your phone, not just an SSH session. See connecting it to Discord for a full setup walkthrough.
  • Configurable terminal backends — choose where the agent actually executes shell commands and code: locally, in Docker, over SSH to another machine, or in a serverless sandbox (Modal, Daytona).
  • A tool pool — web search & extraction, image generation, text-to-speech, speech-to-text, and browser automation, toggled individually during setup.
  • Cron-style scheduling — recurring automated tasks, delivered back to you through whichever messaging platform you’ve connected.
  • MCP integration — connects to any Model Context Protocol server for extended capabilities beyond the built-in tool set.
  • Credential pools — multiple API keys per provider rotate automatically, so a rate-limited or expired key never blocks the agent mid-task.
  • Profiles — run multiple independent Hermes instances on the same machine, each with isolated configs, sessions, skills, and memory. Ideal for separating personal vs. team use, or testing a new model without disrupting your main agent.

Many Surfaces, One Agent

The same Hermes core drives five different interfaces, so you can chat from wherever is most convenient:

  • Interactive CLI (hermes) — the terminal-first chat experience with full streaming and tool execution.
  • Ink TUI (hermes --tui or display.interface: tui in config.yaml) — a terminal UI with docked widget apps (system monitor, clock, kanban board) that you can extend with your own live widgets.
  • Native desktop app (hermes desktop / hermes gui) — an Electron app for macOS, Linux, and Windows with streaming chat, session list, Cmd+K command palette, drag-and-drop file support, and native OS notifications. Extend it with desktop UI plugins that add custom panes and menu commands.
  • Web dashboard (hermes dashboard) — a browser-based admin panel for managing messaging channels, MCP servers, webhooks, memory, and profiles, plus an embedded TUI chat. Secured behind an OAuth or token gate.
  • IDE integration (hermes proxy) — a local OpenAI-compatible API backed by your OAuth provider, so you can point Cursor, Claude Code, Aider, Cline, or any compatible tool at it with no API key of its own.

How It’s Installed

Hermes ships a single-command installer for Linux, macOS, and WSL2:

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

The script installs its own Python 3.11 runtime, uv, Node.js, ripgrep, and ffmpeg, and drops everything into an isolated location under your home directory — no root access needed. After that, hermes setup walks through picking an LLM provider (Nous Portal’s Quick Setup needs only a device-code login, no API key), a default model, which tools to enable, a terminal backend, and optionally a messaging platform.

For a full walkthrough of running it on a free cloud instance, see How to Deploy Hermes Agent on Oracle Cloud’s Free Tier.

Where It Runs

Because a messaging gateway only stays reachable while its process is alive, Hermes needs somewhere to run continuously — closing your laptop shouldn’t take the agent offline. Common choices:

  • A small cloud VPS — including free-tier instances, like Oracle Cloud’s Always Free Arm or Micro shapes.
  • A local machine you leave on — a spare desktop, NUC, or Mac mini.
  • Serverless sandboxes — Modal or Daytona, where the environment hibernates when idle and wakes on demand.

Whichever you choose, the messaging gateway is typically kept alive with a process supervisor (systemd, for instance) so it survives reboots and restarts automatically if it crashes.

Hermes Agent vs. Other Self-Hosted Agents

Hermes isn’t the only open-source, self-hosted personal-agent project — OpenClaw covers similar ground with a mature plugin ecosystem and a web-based Control UI dashboard. The two overlap enough that the choice mostly comes down to what you value more: Hermes’s learning loop, skills library, and wider messaging-channel list, or OpenClaw’s dashboard and plugin maturity. See Hermes Agent vs OpenClaw for a closer look.

Resources

Frequently Asked Questions

What is Hermes Agent?

Hermes Agent is an open-source (MIT licensed) autonomous AI agent from Nous Research that runs on your own server instead of your laptop. It keeps persistent memory across sessions, creates and improves its own skills over time, and can be reached through Telegram, Discord, Slack, WhatsApp, Signal, or a local CLI.

Is Hermes Agent free?

The software itself is free and open source under the MIT license. Running it costs whatever the underlying LLM provider charges (or nothing, if you use Nous Portal's free-tier models), plus whatever server you host it on — which can also be free, on a cloud provider's always-free tier.

Which LLM providers does Hermes Agent support?

Hermes supports 300+ models through Nous Portal, OpenAI, Anthropic, OpenRouter, and custom or self-hosted endpoints. You choose a provider during `hermes setup` and can switch anytime with `hermes model`, without any code changes.

What's the difference between Hermes Agent's tools and its skills?

Tools are built-in capabilities the agent can call directly — web search, code execution, image generation, browser automation. Skills are a larger, file-based library (~160 bundled ones, covering things like GitHub workflows, PDF/Office document handling, research, and coding tasks) that the agent reads and follows, and can also write new ones for itself as it solves problems.

Do I need my own server to run Hermes Agent?

You need somewhere for the agent's process to run continuously, since a messaging gateway only stays reachable while the process is alive. A small cloud VPS works well for this — including free-tier options like Oracle Cloud's Always Free instances.

Can I run multiple Hermes instances?

Yes — profiles let you run multiple independent Hermes instances with isolated configs, sessions, skills, and memory, each on its own model or provider. Switch between them with `hermes profile switch`.

How does Hermes handle multiple API keys?

Hermes maintains credential pools that rotate across multiple API keys automatically, so if one key is rate-limited or expired the agent seamlessly falls back to another without interrupting your workflow.

Related Articles

Side-by-side comparison of Hermes Agent and OpenClaw self-hosted AI agents
AI

Hermes Agent vs OpenClaw: Which Self-Hosted AI Agent to Run

Hermes Agent and OpenClaw both run a self-hosted, always-on AI agent reachable from your messaging apps. Here's how they actually differ.

Four-step diagram for deploying Hermes Agent on Oracle Cloud: SSH key, update Ubuntu, install Hermes, run gateway as a service
AI

How to Deploy Hermes Agent on Oracle Cloud's Free Tier

Provision a free Oracle Cloud Ubuntu instance and install Nous Research's Hermes Agent, a self-improving AI agent that lives on your server 24/7.

Nous Portal unifying model access, Tool Gateway, and Hermes Cloud hosted agent instances under one account
AI

What is Hermes Cloud? Running Hermes Agent Without a Server

Hermes Cloud runs a dedicated Hermes Agent instance for you with no setup. Deploy in minutes, scale to zero when idle, hourly billing, team access.

Troubleshooting diagram for Hermes Agent on a 1GB RAM VPS: PATH, swap, and gateway command fixes
AI

Hermes Agent on a 1GB VPS: Fixing the Common Gotchas

Hermes Agent install issues on a free-tier 1GB VPS: 'command not found' after install, Playwright/Chromium failures, and gateway subcommand mistakes.

All Articles

Free 30-min Production AI consultation

Book Now