Oktopeak

For lawyers who built something that works

Turn your Claude Project into a product lawyers pay for

Your method stays on our server, never in a subscriber's hands. Accounts, Stripe and code in your name. We take a Claude Project, a Claude skill or a custom GPT that works on your laptop and make it something other lawyers log into and pay for.

Get my free Sell-Ready Review

Free, under NDA both ways. A written answer within a week: what the product needs, how many weeks, and what each run costs.

0
lines of your instructions on a subscriber's screen
Yours
cloud, domain, Stripe and code, from the first milestone
6 to 10
weeks to a first version, fixed scope in writing
One person working late at a single lit desk in an otherwise dark office
It works. On one laptop.

What you have today, and what a product needs

The method is the hard part, and you already did it. Everything around it is what lets a stranger pay for it without seeing it.

Your Claude Project todayThe product
Where it runs Your laptop, while it is open and awake A server. A 40-minute job finishes while you sleep and emails the result
Who can use it You Every subscriber, each with their own login
Your instructions In a Project you can open, or a skill anyone you share it with can read On the server only. A subscriber sees the output, never the method
Billing Your own Claude subscription Stripe in your name, per subscriber, with your price
The record Your chat history A log of every run, per customer, for when someone asks what happened
Data terms Whatever your Claude plan says The API under commercial terms, no training on your customers' files, retention you choose

Sharing the Project is handing over the recipe

A Claude skill is a folder of text files, and a shared Project shows its instructions to whoever is in it. A custom GPT looks closed, but it is not. When researchers at Northwestern tested custom GPTs in 2023, they got the system prompt out of 210 of 216 (97.2%), and the uploaded files out of all 24 they tried.

The only way to sell a method without giving it away is to keep it on a server you control and send customers the result. That is most of what the build is.

What happens when you share a Claude skill

97.2%

of custom GPTs gave up their system prompt when asked the right way.

Yu et al., "Assessing Prompt Injection Risks in 200+ Custom GPTs", Northwestern University, 2023. arXiv 2311.11538

The four questions your first buyer asks before the price

Firms buying a legal AI tool ask the same things, in roughly this order. The product should answer them before the call.

1

Do you have a SOC 2 report?

Not on day one, and a serious buyer knows that. What you can hand over: a security packet, the list of every service that touches their data, the model provider's data terms in writing, and a date for a Type 1 audit. We build the product so the audit is paperwork, not a rebuild.

SOC 2 for an early-stage product
2

Will you sign our NDA, and will you sign ours?

Both ways, before you show us a page of your method. Your customers will ask you the same question about their files, so the product ships with the answer: who sees what, where it is stored, and for how long.

3

Who owns the accounts?

You do. Cloud, domain, Stripe, the model provider account and the code repository are opened in your name from the first milestone. If we disappeared tomorrow, any developer could pick it up from your accounts.

4

Can a subscriber see my instructions?

No. The method runs on the server; the subscriber's screen receives the output. Before launch we attack it the way a curious customer would, asking it to reveal its instructions and files, and fix whatever leaks.

What it costs to run, per customer

The number nobody gives you, and the one your subscription price depends on. A method of several hundred pages is around 400,000 tokens. Read in full on every run, that is real money per output. Prompt caching charges the repeated part at a small fraction of the price, and that one setting can decide whether the business works.

The review measures it on your own sample matters, so the price you set is arithmetic.

Work out your own cost per run

One run, a 400,000-token method, Claude Opus 5.5

Read in full every time$1.60
With prompt caching$0.08

Input for the method only, at Anthropic's list price of $4 per million input tokens and $0.20 per million cached reads, checked 28 October 2026. Documents and output come on top.

Three methods, three products

Three shapes this takes. Yours will differ in the details and match in the build.

A drafting rulebook

TodayHundreds of pages of rules a litigator wrote for how motions should read

The productDrafts that follow the rules, sold to other litigators by subscription

An intake method

TodayThe questions an immigration lawyer asks, in the order that works

The productA client questionnaire that turns into a case summary the firm can act on

A memo method

TodayA tax attorney's template and checklist, refined over years

The productFirst-draft memos in the house voice, for small firms without a specialist

This is for you if

  • Your method works in Claude or ChatGPT on your own matters, and other lawyers have asked for it.
  • You want to charge for it without handing over the instructions.
  • You want to own the business: the accounts, the customers and the code.

Stay where you are if

  • It is for you and your own firm. Keep it in Claude; it is free that way.
  • You want your whole firm on Claude with your practice system: that is the Firm Plan.
  • You built it at a firm. Check who owns it before you sell it.

From the review to the first paying subscriber

Every step has written criteria and an exit for you. No price on this page because the honest one depends on the method; you get it in writing after the review.

  1. Week 0

    Sell-Ready Review

    Under NDA. We read the Project or skill and test it on your own sample matters. You get a written answer: what the product needs, how many weeks, and what each run costs at current model prices.

    Free. Keep it either way.
  2. Week 1

    Scope and price in writing

    A fixed scope, a fixed price and a date. Every milestone has acceptance criteria you can read without a developer.

    No hourly billing
  3. Weeks 2 to 9

    Build in milestones

    Logins, the server that keeps your method private, billing, the run log, your branding. Accounts in your name from the first milestone.

    You see it working every week
  4. Launch

    First paying subscribers

    Instructions tested for leaks, costs measured on real runs, terms and privacy page drafted for your lawyer to review.

    The code is yours
  5. After

    A plan you can cancel

    Updates when the model changes, fixes, new features. Cancel it and you keep everything.

    No lock-in

Platforms we built for regulated work

Client names stay masked. The numbers and the screens are real.

The Sell-Ready Review

Tell us what your method does in a line or two. Leave the confidential detail out; that comes after the NDA.

  • An NDA both ways, before we read anything.
  • We run your method on your own sample matters.
  • You get it in writing: what the product needs, the weeks, the cost per run, and a fixed price.
  • Free. If the honest answer is "keep it in Claude", that is the answer.

[ FAQ ]

What founders ask before the review

No. Most lawyers who build these wrote the method in plain English inside Claude or ChatGPT, and some put a website in front of it with a site builder. That is the right way to prove the idea. The product build is our job; your job is the method and the customers.

Only after an NDA signed both ways, and only the people working on your product. We need to read it to build the server around it and to measure what it costs to run. It stays in your accounts, not ours.

No, but most lawyer-built methods today are written for Claude and run best on it. During the review we run your sample matters on the current Claude models and, if it matters to your price, on a cheaper one, and show you the difference in output and cost side by side.

It depends on how big your method is and how big the documents are, and the review gives you the number. As a rule of thumb at Anthropic's list prices on 28 October 2026: a method of several hundred pages is around 400,000 tokens. Read in full on every run by Claude Opus 5.5, that input costs about $1.60 a run; with prompt caching, the repeated part drops to about $0.08. The documents and the output come on top. Your subscription price should come from that arithmetic, not from a guess.

A fixed price, in writing, after the free review, because the honest number depends on the method. Most first versions take six to ten weeks. We do not bill by the hour.

Yes. Keep using it and keep improving it. The product reads the latest version of your instructions, so the work you do in the meantime is not lost.

Selling software to lawyers is a business, but how you market it, who uses it, and whether you share fees can raise questions under your state's rules on fee sharing and unauthorized practice. Ask your bar's ethics counsel before launch. We build the product; we do not give that advice.

We built it for lawyers first, because legal methods are long, precise and valuable. The same build works for any expert whose method lives in a Claude Project or a custom GPT: medical-legal reviewers, consultants, compliance specialists.

Still have questions?

Get my free Sell-Ready Review