Hermes Agent: Your Self-Hosted AI Assistant (Intro + Setup Guide)

Hermes Agent is an open-source AI assistant from Nous Research that runs on your own machine, not as a cloud subscription. Think of it as a personal assistant you install once: it keeps a persistent memory of you across sessions, builds its own library of reusable skills, and can connect to Telegram, Discord, WhatsApp and Slack so you can reach it from your phone.

Its core idea is a learning loop. Every time you finish a task together, Hermes can turn the successful approach into a skill it reuses next time. Over a few weeks it gets noticeably better at the things you actually do, instead of starting from zero in every chat.

ChatGPT is a service you visit; Hermes Agent is software you own. Your chats, files and memory stay on your machine, which matters under Singapore's PDPA if you handle customer or colleague data: nothing leaves your computer unless you connect a cloud model or service yourself. You also choose the brain behind it. Hermes works with 300+ models from providers such as OpenAI, Anthropic, Google and OpenRouter, or fully local models through Ollama.

If you have seen the OpenClaw vs Hermes comparison on this site, the short version is that OpenClaw is a multi-platform gateway while Hermes focuses on learning and memory. This guide covers Hermes on its own, from install to your first scheduled job.

Installation is a single command on Linux, macOS or WSL2 (Windows users run a PowerShell installer):

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

When it finishes, reload your shell, then start it:

source ~/.bashrc

Then type hermes to open the chat. On first run, the fastest setup is hermes setup --portal: one OAuth login that wires up a model plus the tool gateway, which covers web search, image generation, text-to-speech and a cloud browser. The installer keeps everything in ~/.hermes/ on Linux and macOS, or %LOCALAPPDATA%hermes on Windows.

Once Hermes is running, type naturally, like messaging a colleague:

Remember that I run a small F&B business in Singapore and my busy time is Friday to Sunday. Suggest a weekly prep list I can reuse.

Hermes stores what you tell it in persistent memory, so the next session already knows your context. When you solve something non-trivial, ask it to save the approach:

Turn the steps we just used to draft this reply into a skill called customer-reply-drafts, and reuse it whenever I ask for a customer reply.

Skills live as plain files you can read and edit, so nothing is a black box. Type /skills in the CLI to list them.

Hermes has built-in scheduling. Tell it in plain language what you want and when:

Every weekday at 8am, check the weather in Singapore and remind me if rain is expected, plus my first meeting of the day from my calendar file.

To reach Hermes from your phone, connect Telegram. Run /platforms in the CLI, choose Telegram, and follow the steps: create a bot with @BotFather, then paste the bot token that you created into Hermes. Treat that token like a password and keep it private. Once connected, message your bot from your phone to chat, send voice notes and trigger your scheduled jobs remotely.

Type /model to switch brains. Good starting points: the Nous Portal (simplest, one login), a paid API such as OpenAI or Anthropic (best quality, pay per use), OpenRouter (one key, many models), or Ollama for fully local free models on a capable laptop. There is no Hermes subscription fee: the software is open source, so you only pay for the model you choose, and local models cost nothing. If a provider asks for a key, use a placeholder like [your-api-key] in your notes and never paste real keys into chat.

Two Singapore-specific tips. First, PDPA: keep customer or colleague personal data out of prompts sent to cloud providers unless you have consent, exactly as you would with any AI tool. Second, scheduled jobs only run while your machine is on, so many people run Hermes on a small always-on computer or home server; for the general deploy pattern, our OpenClaw deploy guide is a useful companion, and you can browse more agent tools in the tools directory.