By asahifile Hermes Agent is an open-source, self-improving AI agent from Nous Research. Unlike a chatbot that lives in a browser tab, Hermes runs on your own machine — a laptop, a $5 VPS, whatever you have — and executes real tasks: running commands, reading and writing files, browsing the web, and automating your daily workflows. This guide walks you through a complete installation and first conversation in under ten minutes.
What makes Hermes different
Three things set it apart from coding copilots like Claude Code or OpenAI’s Codex. First, it learns from experience: every workflow you complete can be saved as a reusable “skill” that loads automatically next time. Second, it has persistent memory — it remembers who you are, your preferences and environment details across sessions. Third, it is not tied to your keyboard: the same agent can be reached from Telegram, Discord, Slack, WhatsApp, Feishu/Lark and about twenty other messaging platforms while it works in the background.
Step 1: Install it
On Linux, macOS or WSL2, a single command does everything — it bootstraps the package manager, Python, dependencies and the launcher:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bashOn native Windows, run this in PowerShell:
iex (irm https://hermes-agent.nousresearch.com/install.ps1)If you prefer a graphical application, download the Hermes Desktop installer from the official website instead — it bundles the command-line tool and a native desktop UI.
Step 2: First-run setup (one OAuth for everything)
The fastest path to a working agent is the Portal setup, which wires up a model and all four built-in gateway tools (web search, image generation, text-to-speech and a cloud browser) with a single sign-in:
hermes setup --portalYou can also run hermes setup without flags for the classic step-by-step wizard. The wizard writes settings to ~/.hermes/config.yaml and keeps API keys in a separate ~/.hermes/.env file — settings and secrets are deliberately separated, so keep that structure when you edit things by hand.
Step 3: Pick a model
Hermes is provider-agnostic: OpenRouter, Anthropic, OpenAI, Google, DeepSeek, xAI, local models and 20+ other endpoints all work, and you can switch mid-conversation. Run the model picker and follow the prompts:
hermes modelWorried something is misconfigured later? One command diagnoses the whole install:
hermes doctorStep 4: Your first conversation
Launch the interactive chat by simply running hermes. For one-shot queries without entering a session, use the -q flag:
hermes chat -q "Summarize the top AI news today and save it to ~/news.md"Prefer a window and menus? Run hermes desktop for the native app, or hermes dashboard for a browser-based admin panel with an embedded chat.
Where to go next
Once the basics work, the highest-value next steps are: connecting a messaging gateway so you can reach your agent from your phone (Telegram and Feishu/Lark take about ten minutes each), letting it run scheduled jobs with its built-in cron, and reviewing the skills it saves as you work — that growing skill library is where the real leverage accumulates.
Official documentation: hermes-agent.nousresearch.com/docs. Commands verified against the current release. Images: CC-licensed with attribution in captions.