Build Your Own Personal AI Assistant
The same convenience Meta's Muse sells you — errands run, email read, morning brief delivered — except the agent lives on a computer you control, the brain is whichever provider you point it at, and the memory stays in the house. Compiled from the walkthrough in Your Agent, Their Terms (October 2026). Dated so you know exactly what era of the fast-moving AI landscape you're looking at.
Written: October 8, 2026
Covers: what these assistants are, how the parts fit together, and what the setup actually takes. Tools and prices verified as of that date; they drift. When this page smells stale, assume the tools moved on.
Five parts, and who can read each one
Every personal AI assistant — Muse, yours, anyone's — is the same five parts. The only variables that matter are where the brain runs and who can read the transcripts.
You and your phone
Where you type, where answers land. The only part that's unambiguously yours.
The messenger
How your phone talks to the assistant — Telegram, Signal, a web page. Encryption covers specific hops, not the whole path.
The agent
The program that receives your messages and does things — the harness. Yours can run at home, on a computer you control.
The brain
The AI model that writes the answers. Runs on your machine (local) or on rented servers (cloud). This is where the privacy choice actually lives.
The memory
Notes, files, settings, history. If this lives on your hardware, your assistant's past is yours to keep or delete.
The two questions to ask about any assistant: where does the brain run, and who can read the transcripts. If you can't find the answers in ten minutes, that is the answer.
Two sessions, one weekend
The full instructions are guided prompts you paste into any major AI chat — Claude, ChatGPT, or Gemini with web search on. The AI checks whether your machine can run this at all, then walks you through the build one step at a time, teaching as it goes. Nothing about it assumes you've used a terminal before.
The plan
What you want the assistant to do, what it must never see, who else uses the machine, what you're protecting and from whom. You come out with a one-page Build Plan and a working map of where your data goes. You can stop here and keep the knowledge.
The build
Install the agent on your machine, have your first conversation, connect your phone, then lock it down — limited user, no open doors, permissions added one at a time, keys and spending caps where they live. A finish-line test proves what's yours.
Get the complete Part A + Part B prompts
Enter your email and we'll send both prompts, ready to paste — plus a heads-up when the guides are revised as the tools change.
Free. No spam. You'll also be subscribed to the Resilient Tomorrow newsletter.
Your prompts are on the way.
Check your inbox — and if you'd rather not wait, the full walkthrough with the prompts is live on Substack now:
What it costs, and what it takes
Numbered as of October 8, 2026.
An evening to start
Setup is 2–5 hours across the two sessions. Expect to tinker a little more than a normal person should have to. That's the deal.
$0–20/month
The agent is free and open-source. The brain is the cost: local models run free on your hardware; near-frontier cloud brains run about $20/month, less than most single video subscriptions.
The keys
Same convenience, different keys — and a cage you hold the key to isn't a cage. When the terms change or the service shifts under you, your data and your history move with you, not with them.
Credits, and what this method has been tuned on
Mike Jones has run his own personal assistants this way since early 2026 — starting with the open-source OpenClaw project, which proved the thesis (your agents can live at home) but was too high-maintenance to hand to a normal person. The walkthrough in its current form was written for and battle-tested on Hermes Agent, an open-source personal agent from Nous Research that runs on your machine, keeps its memory and credentials vault on your hardware, and connects to whichever AI brain you point it at — local or cloud, terms your call.
Full disclosure in the same spirit the dispatch applies to Meta: Nous Research is the maker of Hermes, and they publish Hermes. Mike runs it, pays for his own cloud brains, and was not paid to write this.
Who Watches the Sentinel?
The organizations doing AI oversight work, the laws already on the books, the federal bills in flight, and the trackers to watch it yourself.
Open the oversight index →Everything AI on this site
The AI hub: every dispatch, guide, and living index in one place, refreshed as new pieces land.
Open the AI index →