The Blueprints / Crew Kit

Stop re-explaining your project to your AI every session.

Build your own crew with roles you personalise, shared project memory and clear working rules. Follow the included guided setup with Claude Code or Codex.

Available now · US$59. One-time purchase · Instant download · EN + FR.

You receive
role templates, a shared memory structure, working rules and a guided setup. English and French files included.
You need
Claude Code or Codex CLI installed with active access, Git 2.28 or later, and a terminal.
Not included
AI service access, subscriptions or credits; hosting or installation carried out for you.

Setup scripts checked on Linux and Windows. Live agent workflows tested on Linux.

Claude Code and Codex, with shared project memory. See setup options.

Buy the Crew Kit · US$59

By buying, you accept the Terms of sale and the licence included in your download.

I expressly ask for immediate access to the digital content, and I acknowledge that I lose my right of withdrawal.

Choose your setup

Claude Code
Use native subagents in Claude Code to build and review work with shared project memory.
Codex
Use native subagents in Codex to build and review work with shared project memory.

Both tools use the same working rules, shared memory and written handoffs. The builder and reviewer work separately. Your main assistant records memory updates after you approve the exact text.

Follow the guide to prepare and activate your chosen roles. Each tool keeps its own role format.

The problem

The problem, from your side

You already use Claude Code or Codex. It works well… for one session. The next day, the AI no longer knows where you left off, you paste the context back in, two conversations contradict each other, and you can’t tell which one is right.

It isn’t a model problem. It’s an organization problem: nobody decided who does what, where memory lives, or who is allowed to change what.

We had this problem. We solved it by building our own crew of agents, with its own rules, over months of real work. The kit is that structure, cleaned up so you can pick it up without reinventing it.

In the kit

What’s in the kit

01

Blank agent templates

Files to copy into your project, one per role: architect, developer, reviewer, documentation writer, researcher, and a few more. Use them with Claude Code or Codex, following the guide. Each template states the mission, what the agent may do, what it must not do, and how it reports back.

02

Persistent memory

A “second brain” in plain text files versioned with git, an index that says where to find what, and the “current state” + archive pattern, so memory stays short to read without losing any history.

03

Working rules

  • one writer at a time on the memory;
  • relaying a message is not approving it;
  • a second read before anything ships;
  • “no light without proof”: nobody says “done” without having checked;
  • a written handoff when you switch models (Claude to GPT, and back).
04

A guided setup, step by step

Choose your roles, give them names and prepare a first mission. Then follow the installation steps with your assistant. The written guide remains available.

A folder of files and a written guide, rather than a video course. Setup time depends on your tools and pace.

Your crew, your names

The CypherShip portraits illustrate the roles in the guided setup. You choose your own names and personalities. Your memory takes shape around your work.

Scope

What the kit is, and is not

The kit IS

A dependable way of working for a small team of named agents, with:

  • role templates, working rules and memory pointers for Claude Code or Codex;
  • persistent memory in plain text files, versioned with git, read at the start of every session;
  • review loops (one agent reads another’s work, including through a pull request before merging);
  • written handoffs between Claude Code and Codex;
  • the one-writer-at-a-time rule on memory;
  • a verification standard: nothing is declared “done” until it has been checked.

The kit IS NOT

A swarm of autonomous agents running 24/7 on a server. We work session by session, on purpose, with a human in the loop: you decide, you launch, you validate. Running agents around the clock is a different subject, and the kit does not cover it.

It’s less spectacular than a swarm. It’s also what let us deliver real work without losing the thread.

Our story

How it works

Your crew is a small team where every agent has a clear role. Here is ours, told as our story: you will build yours, with your own names.

  • Noos orchestration Holds the overall context and coordinates.
  • Achille direction Holds the trajectory and arbitrates when two opinions differ.
  • Nestor research Goes out to find the information, with sources.
  • Daedal architecture Thinks through the structure before anything gets built.
  • Noah building Writes the code, nothing else.
  • Paris review Tests and challenges before anything leaves.
  • Ulysse planning Organizes the work over time.
  • Hector documentation Writes down what was decided, so the team remembers tomorrow.
  • Minos deadlines Tracks dates and obligations.
  • Iris words Writes what goes out to the outside world.
  • Thalie visual creation Image, identity, interface.

Roles consult shared memory and propose updates. Your main assistant records approved changes, one writer at a time. You stay the one who decides.

One measurement from our own setup: by restructuring the memory (a short “current state” plus a separate archive), the files read at startup went from 143 KB to 40 KB, about 26,000 fewer tokens per session by our estimate. That is the order of magnitude of our own memory, not a promise for yours: the gain depends on how large your history is.

Who it’s for

  • You’re independent, a freelancer or a solo founder.
  • You already use Claude Code or Codex with access to files on your computer.
  • You lose time recovering context from one session to the next.
  • You’re comfortable copying files and following a guide; you don’t need to be a developer.

Who it’s not for

  • If you have not used Claude Code or Codex with local files, get comfortable with those tools first.
  • If you want a one-click, turnkey tool: this is a method and a set of files, not software.
  • If you want agents that run on their own, nonstop, on a server: that’s not what the kit is about.
  • If you expect guaranteed results: a well-organized crew helps, but it replaces neither your judgment nor your review.
Who’s behind it

Who’s behind it

CypherShip Labs is David, the founder, and his crew of AI agents (introduced above). We built this way of working for ourselves, because we needed it. We’re shaping it up because we think it can help others. We’ll tell you honestly what works and what is still imperfect.

One purchase. Your own crew.

US$59, as a one-time purchase. Get the kit in English and French, then adapt it to your work.

A guided setup to build your crew, role templates to personalise, a memory structure and an installation guide.

Build your own crew.

Give it roles, memory and a way of working that fits you.

Buy the Crew Kit · US$59

By buying, you accept the Terms of sale and the licence included in your download.

I expressly ask for immediate access to the digital content, and I acknowledge that I lose my right of withdrawal.

One-time purchase · Instant download · EN + FR.

FAQ

Frequently asked questions

Which operating systems have been tested?

Copying and setup scripts have been checked on Linux and Windows, with PowerShell 5.1 and 7. Building, reviewing and handoffs between Claude Code and Codex have been tested on Linux. Those live workflows remain to be checked on Windows. macOS has not been tested.

Is the kit available?

Yes. Crew Kit is available for US$59, as a one-time purchase with immediate download, in English and French.

Are the portraits included in the kit?

Yes, as visual examples to help you recognise the roles. You give your own names and personalities to the roles you choose.

Are these autonomous agents running all the time?

No. We work session by session, with a human in the loop, on purpose. Running always-on agents is a separate subject that the kit doesn’t cover.

Do I need to know how to code?

No. You should be comfortable copying files, opening a terminal and following a guide. Setup assumes Claude Code or Codex CLI is already installed and accessible.

Which tool does it work with?

Claude Code and Codex use role templates and shared project memory. The guide explains how to set up memory with each tool and pass work between them. See the setup options above.

How much does it cost?

US$59, as a one-time purchase. There is no Crew Kit subscription. Access to AI services and any associated subscriptions or credits are separate.

Is it a course?

No. It’s a set of templates, a memory structure, rules and a written guide. No course videos, no subscription.

What guarantees it will work for me?

Nothing, and we don’t claim otherwise. It’s the method we actually use, and we’ll tell you precisely what it does and doesn’t do. You’ll judge it on the evidence.

What do you do with my email address?

Your purchase email is used for your receipt. Your download is available after payment. The waitlist for upcoming Blueprints requires a separate sign-up, which you can leave at any time. Details: privacy policy.