The control system for serious AI work
Mothex is a document-based method that surfaces lost context, silently redone work, and “done” claims nobody verified, and governs them. It runs in plain documents you own, with Claude, ChatGPT, Claude Code, Cursor, or anything you use next.
The name
Prefer it set up for you? Mothex Installation →
The problem
“Claude keeps forgetting what we decided.”
“I spend the first 15 minutes of every session re-explaining my own project.”
“It said it was done. It wasn’t.”
“It rebuilt something that was already finished.”
“My project got too big for one chat, and now I honestly don’t know what’s true.”
This isn’t a prompting problem, and better prompts won’t fix it. Every AI session is a brilliant contractor with total amnesia: highly capable, zero memory of your project, confident enough to claim completion, and happy to rebuild what’s already built. Run one chat and you’ll barely notice. Run five, eight, twelve, and the failures compound: decisions evaporate when a thread fills up, every new session bills you a re-explanation tax, and “done” becomes a feeling instead of a fact.
Here’s the question that exposes the whole thing. Ask yourself, right now: what is actually done on your project? If the honest answer is “I’d have to go check across a bunch of chats”, you don’t have a project. You have nine projects that occasionally agree.
The method
Mothex is not software. It’s an operating discipline that runs in five kinds of plain documents, and it changes four things.
01 READY · ACTIVE · BLOCKED
A Master Roadmap holds every work item with two separate statuses: does it exist (done / partial / absent / unproven) and where it stands (ready / active / blocked / claimed / proved). One glance answers “what’s actually done?”, with receipts.
02 DONE-CLAIMED DONE-PROVED
DONE-CLAIMED is what the AI says. DONE-PROVED is what the evidence shows. Every item declares, in advance, what kind of proof would count: a file, a passing test, a live check, a document. Nothing crosses the line without it. You stop choosing between blind trust and re-checking everything.
03 BOOTSTRAP → HANDOFF
Every session opens with a one-paste Bootstrap that orients it before work starts, and closes with a structured Handoff its successor can stand on. You stop paying the re-explanation tax at the start of every session.
04 CONTROL · DELTA
Exactly one session, yours, the control thread, is allowed to change master truth. Every worker session proposes changes through a structured Delta, stamped with the version it was based on, so stale information is surfaced before it gets merged. Finished work sits behind explicit do-not-redo boundaries the AI is told to respect. Discovered problems go into an Issue Register instead of derailing the thread that found them.
That’s the whole trick: the governance disciplines organizations use to manage human contractors (scoped work orders, evidence requirements, change control, handover documents) compressed into a short daily routine one person can run.
It’s deliberately not software. No installs, no subscriptions, no lock-in, nothing that breaks when your AI tool ships an update. Documents you own, a discipline you keep, with any AI, current or future.
What you get
Do it yourself
$99 USD, one-time
Everything required to set up Mothex on your project this week:
The 11-template system
The Field Manual (~35 pages): the full method, with application notes for Claude Projects, ChatGPT Projects, Claude Code, and Cursor. The 60-Minute Quick-Start. A complete worked example: every artifact filled in for a realistic messy project. Both formats (Markdown pack + Notion workspace) and updates to the tier you bought, for as long as Mothex is published.
Get the Operator Kit — $99Operator Kit Pro · $199 USD adds two more worked-example sets (a content/marketing operation; a research-and-client-work program), the extended Control Thread Playbook, video walkthroughs, and one Setup Review: send your filled Standards Card, Master Roadmap, and first Handoff; get written, annotated corrections back within five business days. Get Pro →
Team License · $249 USD: everything in the Kit for three named users. Get the Team License →
Done for you
$750 USD, flat
You pay a 50% deposit and answer ten questions. Five business days after I have both, you receive your project’s complete governance system, configured around your real, current, messy state: a seeded Master Roadmap with honest statuses, a seeded Issue Register, your Standards Card in your own words, templates tuned to your exact tools, a one-page State Capsule, a written 15-minute daily operating rhythm, a recorded walkthrough of your setup, and one revision round.
A done-for-you setup of a method in documents you own. No access to your accounts, no software, and not engineering or security work.
Ask about an InstallationKeep the adult supervision afterward with the Governance Retainer ($299 USD/mo): two written governance reviews a month (delta audits, stale-state checks, issue triage). Enquire →
It starts free, either way: send two paragraphs about your project and get a written diagnostic: where you’re losing context, what’s probably falsely “done”, and what a Mothex setup would look like for you. No call required.
Get the free diagnosticFrom the creator
I’m not a developer. I started building a large software project using nothing but AI and my own systems thinking. Then it grew, and the real problem appeared: not the code, the coordination. Eight or more sessions running in parallel. Decisions made in thread 4, contradicted in thread 11. A finished module rebuilt from scratch by a fresh session that didn’t know it existed. A component marked done in week two that quietly never worked. I wasn’t building anymore; I was performing archaeology on my own project every morning.
I looked for the system that fixed this. Prompt libraries fix single answers. Project tools assume workers who remember. Agent frameworks are software for developers. Nothing governed the work between sessions, so, failure by failure, I built it. A truth ledger. A split between claimed and proved. Handoffs that actually transfer context. One thread that controls, while the rest propose. Each mechanism exists because its absence cost me real days, and I can tell you which days.
Mothex is that system, generalized so anyone can run it, in their own documents, with whatever AI they already use. I publish it under a working name rather than my own because the method should win on evidence, not on a face. The kits are refundable for 14 days after you buy.
— Mateus Ventura, creator of Mothex
Pricing · all prices in USD
$99
USD, one-time
$199
USD, one-time
$750
USD, flat
Rule of thumb: running four threads or fewer and happy to read a manual → Kit. Want certainty you set it up right → Pro. Running five or more threads, already burned, no time → Installation. Team of three? → Team License, $249 USD. Ongoing governance reviews after installation: $299 USD/mo, cancel anytime. Kits: 14-day full refund, no questions.
* Updates to the tier you bought, for as long as Mothex is published.
FAQ
No, and that’s deliberate. Mothex is a method that runs in documents you own. No software to install, nothing that breaks when your AI tool updates, nothing to be locked into. (The done-for-you “Installation” means I set the documents up for you, in writing. I never touch your accounts.)
Anyone running serious work across three or more AI sessions: founders building products, developers running coding agents, consultants delivering client work. Who is it not for? If you use one or two chats casually, you don’t need this yet. Bookmark it for the week that changes.
Memory reduces one cost, re-explaining, inside one tool. It doesn’t coordinate parallel sessions, doesn’t verify “done” claims, doesn’t stop rebuilds of finished work, and doesn’t follow you across tools. Mothex governs all of that, above any single tool’s memory.
Those are good free frameworks for how coding agents build software: roles, specs, pipelines. Mothex governs what’s true across your whole program: the evidence ledger, claimed-vs-proved, handoffs, your control position. Different layer; you can run Mothex above them. And it works for non-code work too.
You could, the way you could write your own accounting system. The value isn’t the idea of “keep a ledger”; it’s the failure-tested specifics: what a handoff must contain so the next session actually stands on it, how to catch a stale update before it corrupts your plan, where “done” claims slip through.
Fair question, answered with inspection rather than credentials. The worked example and sample templates are public. The method’s depth is the résumé. There’s a 14-day full refund on the kits. And the free diagnostic means you can watch the thinking applied to your project before spending anything.
About 60 minutes following the Quick-Start: fill your Standards Card, seed your roadmap with 5–15 items, run your first bootstrapped session, close it with your first handoff. The first week of running it is where the habit forms.
No. If you can maintain a checklist, you can run Mothex. The two ledgers are tables; the templates are fill-in documents; the manual explains every term in plain language.
Yes: research programs, content operations, courses, client work. Pro includes worked examples for exactly those shapes.
Kits: fourteen days, full refund, no questions. Ask through your receipt or email me. The Installation is guaranteed on deliverables: if the eight scoped items aren’t delivered, you get every dollar back; before work starts, the deposit is fully refundable. Refund policy
Every week without a control layer costs the same things: the re-explanation tax, the silent rebuild, the “done” that wasn’t. Every week with one compounds the opposite.
