Install Hermes Agent on my server and connect it to my Telegram bot.
Copy the prompt
Each lesson hands you one, shaped by your own choices.
Build your own personal AI assistant, one guided step at a time. It is self-hosted on your own setup and shaped around your life and your work.
One AI agent for your life and your work, inside the apps you already use.
Self-hosting your apps? Your agent runs on the same network, so it reaches them directly.
Works with Claude Code, Codex or Cursor. Every lesson ends with something you can see working.
Install Hermes Agent on my server and connect it to my Telegram bot.
Each lesson hands you one, shaped by your own choices.
Claude Code, Codex or Cursor does the typing.
A test you run yourself closes every lesson.


The path shows you how to install Hermes Agent or OpenClaw, then goes well beyond their docs: the tools, tuning and extras that make an assistant dependable.
They run on a vendor's service. Yours runs where you decide.

Hosted assistants as described in public reviews, November 2026.

Now it is the first place I go.
I tried the hosted assistants first. One ran on a small cloud machine that ran out of room, and it kept doing things I never asked for. Another had a security hole.
None of them could reach the apps I host myself. So I built my own, gave it every tool and tuned it fast. It learned how I work.
Six everyday jobs, from the request to the finished result.
Plan my day around the dentist.
Six things, two quiet hours, home by six.
Reply to the landlord. Keep it polite.
Re: Boiler repair
Hi Mr. Miller, Tuesday at 10 works for the repair.
Remind me to call Mom at six.
Four days in Lisbon in May.
Movie night.
Is everything up?
Checking five servicesAll up. Last backup at 03:00.
60 ready requests, each with the exact message to send.

From a blueprint to your agent, running. Every lesson ends in something you can see working.

Designed, not yet running.
Framework, model and home chosen
Awake, and talking to you.
The first message on your phone
Safe to rely on.
Rules, a tested backup, an alarm
Useful every day.
Tools, routines, memory, skills
Part of the household, or the business.
Your people in, a monthly checkMost people get an agent running and then wonder what to ask of it. The path comes with a catalog of things to try: ready requests, skills and setups, many of them lifted straight from the agent we run ourselves. We keep adding to it as the tools change.


The exact message to send your agent, for both frameworks.

Procedures it follows, in plain text you read before you install.

Configurations and add-ons, each with a short guide and a prompt for your coding agent.
Nothing in the catalog is required. Pick what you would use, and add the next one when the first has earned its place.
You decide what it may do alone, what needs your yes, and what it may never do.

An agent that reads your mail and acts for you has to be built carefully. Every stage adds a protection, and you prove each one works before the agent gets more to do.
No agent is safe by design, and the path says so from the start. So yours is built for the day it gets tricked: little within its reach, a yes from you before anything lasting, and a way back.
12 rulesThe safety guide comes with the path, and we keep it up to dateFour things. We teach and guide you through every one of them.
A home server, a Raspberry Pi, or a small server you rent.
With a hard spending cap. For one light user:
Your coding agent does the typing.
Where you talk to your agent. Pick the one you already use.

Your coding agent does the typing. You make the choices and check the result.
An AI assistant that runs on a machine you control, such as a home server, a Raspberry Pi or a small server you rent, instead of on a vendor's service. You pick the model, its memory is files on your own disk, and you talk to it in your chat app.
A hosted assistant runs on its vendor's service, with that vendor's models and the apps they chose to connect. Your own agent runs where you decide, uses any model you pick, and reaches your own apps directly.
You can. Hermes Agent and OpenClaw both install quickly from their docs, and the path covers that too. What it adds is everything around the install: the add-ons, tools and tuning we tested, improved and collected.
A machine that stays on or a cloud VPS, a model account, a coding agent and a chat app on your phone. The path guides you through setting up each one.
Hermes Agent and OpenClaw are open source. You pay for the machine, about $0 to $15 a month, and for the model, about $0 to $20 a month for one light user with a hard spending cap.
The path recommends Hermes Agent and fully supports OpenClaw. You can switch later.
Any model you pick. The path walks you through the choices: a subscription sign-in, an API key with a spending cap, a router that reaches many models, or later a local model on your own hardware.
Yes. You talk to your agent in the chat app you already use: Telegram, Discord, Slack or WhatsApp. WhatsApp needs a separate number.
No agent is safe by design, so the path builds yours for the day it gets tricked: its own machine, nothing on the public internet, a short list of who may talk, your yes before anything lasting, and a backup you have restored. A safety guide of twelve rules comes with it.
Its memory is files on your own disk, and it sits on a private network with nothing open to the public internet. What you ask goes to the model provider you chose, unless you run a local model.
A Raspberry Pi or a small rented server is enough. If you have a home lab, your agent lives on it.
Soon. Join the waitlist and you get one email the day it opens.
Name it, join the waitlist, and get one email the day Agent at Home opens.
