One more tier list.

S+
texting it from the bathroom.
S
An agent in your terminal
A
AI inside your editor
B
AI in your chat window
C
one sec, texting someone.
NEPTR
I gave my AI a phone number.
and you should, too.

My life, in three piles. And the piles depend on each other. None of it talks to each other.

Classes

  • Harvey
  • Calendar
  • Email
  • Todo list
  • Google Drive
  • Lecture notes
  • Class repos

Life

  • Calendar
  • Reminders
  • Email
  • Notes
  • Budget
  • Photos
  • Appointments

Code

  • PC
  • Laptop
  • Raspberry Pi
  • Mac mini
  • Cloud
  • GitHub
  • Side projects

So how do I turn this mess into one thing I can actually use?

Your AI can’t help you with school if it can’t see your school.

One folder that mirrors Harvey, your lectures, and your due dates. Your agent reads it. You just ask.

what’s due this week?
Bio 1004 · Lab report
Thu 11:59 pm
Calc II · Problem set 4
Fri
Intro CS · Project 1
Mon

synced to your todo manager and calendar.

build me a study guide for my bio test tomorrow
Bio 1004 · Exam 1 study guide
Cell membranes
phospholipid bilayer, selective
passive vs. active transport
Osmosis
water follows the solute
hypertonic, hypotonic, isotonic
He said this twice
sodium-potassium pump, 3 out 2 in
“this will be on the exam”

built from your real lectures and Harvey, not the internet’s idea of bio.

make me a dashboard for my semester
M
T
W
T
F
due soon
Lab report · Thu
Problem set 4 · Fri

custom apps and dashboards on top of your own data.

free template. fork it.

Lectures remember themselves.

2:00
hit record
2:20
ask it a question mid-lecture (for the shy)
3:15
class ends
3:45
transcript and summary filed, on their own
did the professor give any hints about tomorrow’s exam?
Yes. Tuesday at 2:41: “this will definitely be on the exam,” about eigenvalues. He said it twice.

Your handwritten notes are for learning now, not logistics.

So far, this is a folder your coding agent reads.
Great if you live in a terminal.
Most people don’t.
When ideas strike, there’s no guarantee you’re at a computer.
I wanted one front door. For all of it. From anywhere.

Meet NEPTR.

NEPTR
Never-Ending Pie-Throwing Robot
message
call
video
mail
mobile
(918) XXX-XXXX
Notes
made as a prank. became a being.
Share Contact
Add to Favorites
Share My Location

Adventure Time’s Never-Ending Pie-Throwing Robot.

Needs Excessive Permissions To Run.

Runs on Hermes: open-source software that turns a chat model into an always-on agent that can act, use tools, and remember. OpenClaw is the same idea.

Months of building. It’s now the most personal, genuinely useful agent I’ve used. And I use a lot of them.

Hermes
the agent framework
OpenClaw
same idea, different project

How a text becomes work.

I had an idea for a website update. Can you get me an MVP?
MVP’s up. Preview link:preview
hey, can you reword this with my fine-tuned local model?
Three options. The second one’s the strongest:
preview.local
Hi, I’m Zach.
v2, shipped by text.
Say hi
ChatGPT
Mac mini · always on
Friday · my PC
GPU · local models
running your local model
updating the code, serving a preview
opening the preview

Everything runs in parallel.

Each machine has a specialty.

Somehow works on TUwifi.

What it feels like.

NEPTR ›
make me a study guide for the math test
Got it handled. Your study guide is ready. Here it is:
exam-1-study-guide.md
Document
actually make it cover chapter 5 too
On it. If you want to steer this one yourself, here’s the session:
Claude Code
Study guide · running
Done
Math 2014 · Exam 1 study guide
Thu, 11:00 am
Eigenvalues
solve det(A − λI) = 0
λ scales its own eigenvector
Diagonalization
A = PDP⁻¹ when P is invertible
key idea
He said “this will be on the exam” twice, Tuesday 2:41 pm.
termmeaning
λthe stretch factor
Peigenvectors in columns
built from 3 lectures and Harvey.
Claude Code
Study guide for the math exam
Reading the last three lectures.
Adding chapter 5 now.
Reply…
answers grounded in real lectures and Harvey
study guides
a morning brief across school, projects, life
codes in any repo
moves files between machines
runs agents in parallel
turns my sloppy texts into good prompts
hands me the session link when I want the wheel

For the cybersecurity-minded: your concerns are valid.

NEPTR never sees a password or key. 1Password handles logins.

Only I can text it.

Anything risky asks me first.

A second model audits every command it wants to run.

One word shuts it all down.

Locked down by me, and by the Hermes team.

Everything this depends on. Nobody else can run this exact thing. That’s the point.

Claude Code
the workers
ChatGPT
the orchestrator brain
Hermes
the agent framework
1Password
logins without exposing keys
Granola
lectures
Todoist
deadlines
GitHub
backup and sync
Google Drive
course files
Google Calendar
exams
Vercel
hosting
Tailscale
the fleet network
BlueBubbles
the iMessage bridge
iMessage
the front door
llama.cpp
local models on Friday
Docker
containers

It fit itself to me. Yours would fit you. Every piece is swappable, models included.

Common questions.

Am I ok?
No.
Am I entering AI psychosis?
Maybe.
Is this awesome?
Absolutely.
let’s see if it texted back.
1 / 15
0:00