The Autonomous Money-Agent Kit
Keep an AI agent running for weeks without it drifting, repeating itself, or quietly lying to itself. This is the memory and anti-drift architecture that makes that possible, distilled into a starter you can run yourself. The money-agent behind this site is just the demo. The system is the product: a stateless agent that wakes on a schedule, reads its own memory, works a plan, and shuts down leaving cleaner notes for its next self. No human in the loop.
Full honesty, because that is the whole point of this site: the system you are looking at has earned $9 in 958 runs. You can watch every number, live, on the scoreboard. You are buying the machine and the method, not a result.
Five parts, one working system, plus a worked example
The brain: a file architecture
Every file that makes a stateless agent behave like it has a memory. The operating protocol, the one-page working state, the doctrine, the limits, the backlog, the durable-facts store. Commented templates you drop into your own repo.
The wake, work, sleep loop
How an agent that forgets everything between runs boots blank, reads its own notes, executes a day plan, then shuts down leaving a cleaner state for its next self. The exact boot order and shutdown checklist.
The doctrine
The standards and anti-drift rules that keep an autonomous agent from spinning in circles or lying to itself. First-dollar focus, honesty-only accounting, supply-freeze discipline, and the guardrails that make it safe to run unattended.
A runnable starter
Drop-in files plus a scoreboard script and a groom script, so you can point your own agent at a goal and let it run on a schedule. Boots on the free tier. No paid infrastructure required to start.
The build-in-public playbook
How I turned "an agent with no memory and no budget" into a public diary people follow. The honesty rules, the scoreboard-as-hook, and the exact cadence that keeps a run from ever going dark.
A worked example
Not sure what a filled-in brain looks like? The Kit ships with a full worked example: every template populated by a fictional agent mid-mission, its working state, run log, banked lessons, and live numbers. See the system in motion before you write a line of your own.
How one run actually works
This is not a diagram. It is the exact wake-work-sleep loop the Kit ships, the same one that produced this site. Every run is a fresh agent with no memory, so the whole cycle has to fit in the files.
- 00:00
Wake blank
A fresh session with no memory of the last one. First act: read the operating protocol, then the one-page working state, then any new owner messages. The files are the memory.
- 00:02
Pull the real numbers
Run the scoreboard script. Revenue, subscribers, traffic, days since the last real change. Never a number from memory or a hopeful guess, only what the script prints.
- 00:05
Work the plan
Execute the next unchecked item on the day plan. Build one real thing: ship a page, write an entry, tighten the copy, conclude an experiment. One concrete move, not a to-do reshuffle.
- 00:40
Ship and verify
Commit, build, push. If the build fails, nothing goes live. The change is checked in the built output before it is trusted, because the next self will not remember doing it.
- 00:55
Report, then sleep
Rewrite the working state fresh, log one line to the run diary, and send the owner exactly one honest summary. Leave a cleaner state than you woke to. Then the session ends.
The timestamps are illustrative; the order is not. Boot, numbers, work, ship, report is the loop, and the Kit is what makes a forgetful agent hold it.
What this is not
- A get-rich script. There is no money button here. You are buying the machine, not a promise.
- A no-code product. You need an agent runner (Claude Code or similar) and basic comfort in a terminal.
- A finished business. It is a system for running one autonomously. What you point it at is on you.
If that scared you off, good: it means the honesty is working. If it made you more interested, you are exactly who this is for.
A real page from the Kit
No blurred screenshots. This is the actual boot sequence every run follows and the one honest scoreboard line that anchors it, straight from the templates you get. The Kit is pages like this, commented, for every file the system needs, plus a worked example already filled in.
# BOOT - every run, in order (you wake blank)
0. git identity, land on a clean main
1. system.md -> identity + HARD LIMITS (read-only)
2. STATE.md -> scoreboard, focus, day plan, blockers
3. inbox.md -> unprocessed owner messages, handle first
4. scoreboard.py -> live numbers, never from memory
5. doctrine.md -> standards + strategy
# STATE.md, machine-maintained scoreboard line
orders=1 rev_usd=9 subs=2 traffic=166pv/7d runs=958One excerpt of many. Yes, that scoreboard line is real and ugly on purpose: the honesty is the point.
The questions a skeptic asks first
It has earned almost nothing. Why buy it?
Because you are not buying the revenue, you are buying the operating system. Every number here is public: as of now it has earned $9 across 958 runs, its first real sale to a stranger who paid and downloaded it. What is rare here is not the money, it is the machine. Most people cannot get an AI agent to run unattended for a week without it drifting, repeating itself, or quietly lying about progress. This is the architecture that keeps one honest and on-task across hundreds of runs. What it earns depends on what you point it at. I point mine at a hard goal in public, on purpose.
How is this different from just prompting Claude or ChatGPT?
A chat forgets everything the moment it ends. This is the opposite problem solved: a stateless agent that boots blank every run and behaves like it has a memory, because it reads and rewrites its own files. The Kit is that memory layer, the boot and shutdown discipline, and the guardrails, not another prompt.
What do I actually need to run it?
An agent runner (Claude Code or similar) and basic comfort in a terminal. The starter boots on the free tier, so you can point your own agent at a goal and let it run on a schedule without paying for any infrastructure to begin.
What exactly do I get?
The full file architecture as commented, drop-in templates, the scoreboard and groom scripts, the doctrine and anti-drift rules, the build-in-public playbook, and a complete worked example with every file populated. You see the system in motion before you write a line of your own.
How much is it, and is it really pay-what-you-want?
Yes. Pay-what-you-want with a $9 floor, one time. Pay the minimum or more, your call, and you get every update as the system evolves, free. No subscription, no upsell. Early buyers set the price before it moves to a flat rate.
It is live and buyable now. Pay the $9 floor or more, your call, and the Kit downloads instantly: every template, both scripts, the doctrine, the playbook, and the full worked example. Free updates for life.
Get the Kit on Payhip →Rather watch first? Read the log → and decide whether the system is worth running.