Hermes Agent: the AI agent that learns with you

AI agents that live in your terminal have become a commodity — but an agent that learns from its own experience is still rare. Hermes Agent, from Nous Research, is exactly that: an open source agent (MIT, written in Python) with a built-in learning cycle — it creates skills from the tasks it performs, improves those skills through use, keeps memory between sessions and builds a model of who you are over time. And it isn't tied to your laptop: it runs on a 5-dollar VPS and you talk to it through Telegram while it works in a cloud VM.

What Hermes Agent is

Hermes is a general-purpose AI agent with two entry points: a full terminal TUI (multiline editing, command autocomplete, history, streaming tool output) and a messaging gateway that connects it to Telegram, Discord, Slack, WhatsApp, Signal and email — all in one process, with conversation continuity across platforms and even voice message transcription.

What sets it apart from competitors is its closed learning loop:

• Memory curated by the agent itself — it decides what's worth persisting and periodically “nudges” itself not to forget
• Autonomous skill creation — after a complex task, it records the procedure as a reusable skill (compatible with the open standard agentskills.io)
• Searching its own past — full-text search (FTS5) in previous conversations with LLM summarization
• Native cron scheduler — daily reports, nightly backups and weekly audits described in natural language, delivered on any connected platform
• Parallel subagents — delegates isolated workstreams and executes Python scripts that call tools through RPC

No model lock-in: works with Nous Portal, OpenRouter, OpenAI or your own endpoint (a local llama-server, for example). Switching models takes a hermes model — no code changes.

What you need

• Ubuntu (or any Linux, macOS, WSL2 and even Termux on Android) — here we'll use Ubuntu 24.04
• curl and internet access — the installer handles the rest: uv, Python 3.11, Node.js, ripgrep and ffmpeg
• An API key from an LLM provider (OpenRouter, OpenAI…) or a subscription to Nous Portal
• Modest hardware: a cheap VPS is enough — the heavy work stays with the model provider

Installing on Ubuntu

One line:

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

When it finishes, reload the shell and launch the agent:

source ~/.bashrc
hermes

Everything lives in ~/.hermes. If something seems odd, the agent diagnoses itself:

hermes doctor
hermes update   # atualizar para a versão mais recente

Basic configuration

The fast route is the full wizard, which configures the provider, model, tools and gateway at once:

hermes setup

Or in parts:

hermes model        # escolher provedor e modelo de LLM
hermes tools        # ligar/desligar ferramentas
hermes config set   # ajustar valores individuais
hermes config get   # conferir a configuração

If you'd rather not collect five API keys (model, web search, image generation, TTS, cloud browser), Nous Portal covers everything in one subscription, with more than 300 models available:

hermes setup --portal   # login OAuth + Tool Gateway numa tacada
hermes portal info      # ver o que ficou habilitado

Messaging integrations

The everyday magic is in the gateway: you talk to the same agent from the terminal, Telegram or Slack, and the conversation picks up where it left off:

hermes gateway setup   # assistente: escolhe plataformas e tokens de bot
hermes gateway start   # sobe o gateway (Telegram, Discord, Slack, WhatsApp, Signal, e-mail)

Inside any conversation (CLI or messenger), the commands are the same: /new restarts, /model switches the model, /personality changes the persona, /compress compresses the context, /skills lists learned skills. There's also DM pairing and command approval as security layers.

Other integrations

• MCP (Model Context Protocol) — connect any MCP server and Hermes gains its tools
• Seven terminal backends — local, Docker, SSH, Singularity, Modal, Daytona and Vercel Sandbox; on serverless backends (Modal/Daytona) the environment hibernates when idle and costs almost nothing between sessions
• Home Assistant — home automation through the gateway
• 40+ native tools — shell, files, web search, image, voice, browser, organized into toolsets you enable by usage profile
• Context files — your project's AGENTS.md becomes persistent conversation context

Is it worth it?

If you already use an AI agent in the terminal, Hermes is in another category because of memory and skills: it gets better in your environment as the weeks pass, and the messaging gateway turns any Linux server into an assistant available in your pocket. The project is MIT, has complete documentation and an active community on the Nous Research Discord.

Install it, run a hermes setup and give it a week of real tasks — then tell me on the contact page how it went.