A build guide · field tool Nº 002
Inbox Operator
A step-by-step build guide. In about 90 minutes, with Claude Code doing the typing, you build a private AI email assistant that reads your Gmail, learns how you actually write, and drafts replies inside the conversation, in your voice. It drafts. It never sends. You always do.
no coding needed. you paste prompts, Claude writes and tests the code.
Every draft waits for you. That's not a setting. It's how it's built.
Nº 01 · the problem
Your inbox is a
second job.
The subscription
never clocks out.
Email eats the day in small bites: the same handful of replies, over and over, in a dozen tiny variations. So you either answer everything yourself late at night, or you pay a subscription forever for an AI email tool that drafts in nobody's voice but its own.
Generic AI writes generic replies. It doesn't know your prices, your policies, or the fact you never open a reply with "I hope this finds you well." Inbox Operator learns to sound like you, because it's built on your email, not a template.
what AI-assisted email costs, per month
approximate street pricing for AI-assisted email tools, monthly. Inbox Operator runs on your own metered Anthropic key, capped by you, not billed as a flat seat fee.
Nº 02 · how it works
Four things
it learns. One
thing it never does.
The guide builds this in order, phase by phase, checkpoint by checkpoint. By the end, your inbox has a quiet assistant working inside it.
Move 01
Learns your voice
It reads through your sent mail and builds a private profile: your tone, your greetings, your sign-offs, how formal you get with clients versus friends. Yours to read and correct before anything else happens.
Move 02
Learns what deserves a reply
It studies which emails you actually answer and which you ignore, then shows you a "likely culprit" list of senders and types it thinks should never get a reply. You approve, remove, or add to it.
Move 03
Learns what it needs to know
It mines your past replies for the questions people ask over and over, pulls out your pricing, policies, and links, and asks you to fill in anything it couldn't find on its own.
Move 04
Drafts, never sends
New mail gets classified. Reply-worthy email becomes a draft inside that Gmail thread, labeled so you can find it. Sensitive threads get flagged "Needs You" instead. The send permission is never requested.
safety is the constraint this is built under, not around. no send function exists in the code, at any phase of the build.
Nº 03 · the math
Napkin math,
real cost.
Nothing here is a growth-hack number. It's a 90-minute build and a metered API key with a hard cap you set before the first draft ever gets written.
worked example · one real inbox
Renting vs. owning.
A subscription AI email tool runs $24 to $30 a month, forever, whether you check email twice a day or forty times. A year in, you've paid $300 or more and you still don't own anything.
Inbox Operator costs $19 once to build, plus roughly $3 to $15 a month you control through your own API key's spending cap. First year, you're likely still ahead. Every year after, it isn't close.
estimated cost · same job, two ways, per year
estimate: $30/mo subscription vs $19 once plus roughly $9/mo, the midpoint of the $3–15 range, on your own capped api key. your mix will vary; the cap is always yours to set.
Rent an inbox tool and the bill never stops. Build your own and the only recurring cost is the actual API calls, capped by you, visible any time at console.anthropic.com. That's not a discount. That's owning the thing. the build, not the subscription
Nº 04 · the safety rule
It drafts.
It never sends.
The send permission is never requested and no send function exists in the code, by design, from the very first prompt in the guide. The worst a bad draft can do is sit in your drafts folder until you delete it.
Sensitive threads, legal, financial, medical, anything with real money or real feelings on the line, get flagged "Needs You" instead of drafted. If a reply needs a fact the app doesn't have, it says so and flags it too, instead of guessing.
Read it, edit it, or delete it. The send button stays yours, every time.
Nº 05 · in the box
One guide.
Seven prompts.
A working inbox.
- The full build guide, step by step, from installing Claude Code through a live, learning inbox assistant.
- 7 copy-paste prompts. You paste them into Claude Code in order; Claude writes, tests, and fixes the app for you.
- About 90 minutes end to end, most of it watching Claude work and answering its questions about how you write.
- No coding experience needed. When something breaks, you paste the error back to Claude and say "fix this."
- A checkpoint after every phase, so you know exactly what "working" looks like before you move on.
It's a guide, not a hosted service. Nothing of yours passes through our servers, ever.
Your email, your voice profile, and your knowledge base live in a folder on your own computer. The only outside call is to the Anthropic API you already control. Built and tested by Paul Von Rieter across the same build discipline behind the tools on the operator site: tEE, CRUMB, WERKS, GainSail, and the rest of the arsenal.
the offer
Nineteen dollars.
Own it outright.
- Instant access: checkout lands you on the private build guide
- Build time: about 90 minutes, mostly watching Claude work
- Runs on your own Anthropic API key, with a hard spending cap you set
- Yours forever. No renewal, no seat license, ever.
field tool Nº 002 · digital delivery
Inbox Operator
secure card checkout by stripe. you'll be redirected to the build guide the moment payment clears.
Nº 06 · fair questions