I gave my AI a phone number

On Friday I gave a talk to the AIML Club at the University of Tulsa, where I’m the vice president, about the agent I text. It’s called NEPTR, it has its own phone number, and for the last few weeks it has been the front door to most of what I do. The slides are here if you want the ten-minute version.

The problem

My life is in three piles. Classes, which means lectures, due dates, exams, and Harvey (what we call Blackboard at TU). Life, which is my calendar, my notes, and the ideas that show up while I’m nowhere near a computer. Code, which is a handful of repos spread across three machines. The piles depend on each other, and none of it talks to each other. I wanted one thing I could actually use, from anywhere, without opening a terminal.

The folder my agent reads

The first half of the answer is a repo I think of as my course brain. Every lecture gets recorded, transcribed, and filed on its own about half an hour after class ends. Every assignment and due date is mirrored from Harvey. My agent reads that folder, so I can ask what the professor said about the exam and get the moment in the lecture where he said it. My handwritten notes are for learning now, not logistics.

That part is public. I stripped my own classes out and published it as a free template for other TU students. Fork it.

But a folder your coding agent reads is only great if you live in a terminal, and most people don’t. When ideas strike, there’s no guarantee I’m at a computer. I wanted one front door for all of it.

Meet NEPTR

NEPTR is named after the pie-throwing robot from Adventure Time, the one that was made as a prank and became a being. In my house it stands for Needs Excessive Permissions To Run. He has his own Apple ID, which is a sentence I did not expect to say in public, so he shows up in Messages as a real contact and I text him like anyone else.

He runs on Hermes, open-source software from Nous Research that turns a chat model into an always-on agent that can act, use tools, and remember. I didn’t write Hermes. What I built is everything around it, like the iMessage bridge on an always-on Mac mini, the plumbing that lets him hand work to Claude Code sessions on whichever of my machines fits the job, and the guardrails. (OpenClaw is the same idea, different project, if you want to compare.)

What I actually ask him for is answers grounded in my real lectures and Harvey, study guides, code in any repo, moving files between machines, running agents in parallel, and a lot more. The part I find most useful is that he turns my sloppy texts into good prompts. I throw lazy half-sentences at him and he reorganizes them into a proper prompt for each session. And when I want the wheel, he hands me the session link and I take over.

How a text becomes work

Three machines. A Mac mini that is always on and receives the texts. My PC with a GPU, named Friday, for local models. My laptop. When a text needs a local model, the mini routes it to Friday. When it needs a build or a preview, the mini runs it and my laptop opens the result. Everything runs in parallel, each machine has a specialty, and somehow it works on TUwifi.

For the cybersecurity-minded

Your concerns are valid. NEPTR never sees a password or key, because 1Password handles the logins. Only I can text him. Anything risky asks me first. A second model audits every command he wants to run. One word shuts it all down. Locked down by me, and by the Hermes team.

Nobody else can run this exact thing

That’s the point. The stack is fifteen things I already used every day, which are Claude Code, ChatGPT, Hermes, 1Password, Granola, Todoist, GitHub, Google Drive, Google Calendar, Vercel, Tailscale, BlueBubbles, iMessage, llama.cpp, and Docker. None of it is software I wrote from scratch. Agents wired together the tools I already had. It fit itself to me. Yours would fit you, and every piece is swappable, models included.

One slide near the end has three questions on it. Am I ok? No. Am I entering AI psychosis? Maybe. Is this awesome? Absolutely.

The last slide

The talk opened with me texting NEPTR from the stage, asking him to build the last slide of the presentation, something good, and to text me when it was done. Then I gave the talk, and at the end I refreshed the deck to see what he had made. The slide he wrote is hosted with the rest of the deck, signed in his own voice.