Open-source signal-to-product system

Signal in.
Merch out.

Codex Merch is a hackable pipeline that turns a trusted trend into an original, production-ready garment—not just an image. Model taste, software guarantees, and human release authority stay deliberately separate.

One-sentence Build Week input

Create a preview merch for the trend ‘The Sol Shines’.

Output / 6 panels · catalog entry · hashed proof

Current deployment

Production mode

Secure live payment runs through Stripe. Paid orders are verified by signed webhooks, stored durably, and handed to the fulfillment workflow for a final production review.

Live payment · Real order

The complete loop

Five moves. One inspectable run.

Every stage leaves a contract, artifact, or decision that can be reviewed and replaced. That is what makes the pipeline useful beyond this storefront.

Direct pathOwner-supplied trendNo discovery claim
Weekly path30 authorized X postsRecurring trend or no_trend

One normalized trend contract

  1. 01

    Signal

    Start with a premise

    Use a direct owner brief, or let the weekly workflow inspect exactly 30 posts from an authorized X list. Provenance stays attached, so an owner idea is never passed off as discovered evidence.

    Trusted trend contract
  2. 02

    Direct

    Choose a visual world

    GPT-5.6 proposes exactly three materially different, panel-aware garment systems in strongest-first order. On the weekly route, a weak or unsafe signal ends successfully as no_trend.

    Three ranked recipes
  3. 03

    Build

    Make the real files

    Deterministic code composes the front, back, both sleeves, label panel, and inside label against the provider template. The same inputs reproduce the same files and hashes.

    Six production panels
  4. 04

    Prove

    Review what was rendered

    GPT-5.6 critiques the actual garment output. Code separately enforces rights rules, placement coverage, exact dimensions, PNG integrity, prepress, and repository checks.

    Inspectable proof
  5. 05

    Release

    Stop safely—or ship

    A preview run stops before payment or fulfillment. A production run opens only after separate human approval, signed catalog checks, Stripe configuration, and provider readiness all pass.

    Preview or gated product

Clear authority

Taste is not permission.

The model can make subjective calls. It cannot invent provenance, waive a production check, publish a product, or create an order.

Taste

GPT-5.6 proposes

  • Interpret a recurring signal
  • Rank three garment directions
  • Critique the actual render
Guarantees

Code proves

  • Track provenance and rights checks
  • Render exact provider-sized panels
  • Gate hashes, retries, and publication
Authority

A human releases

  • Authorize the signal source
  • Approve brand and commercial rights
  • Enable providers and commerce

Open by design

Fork the pipeline, not the promise.

The prompts, schemas, renderer, state machine, tests, and adapters live together in the repository. Each seam can evolve without turning the whole system into a black box.

01 / Signal

Bring another source

Replace X with a community feed, search signal, sell-through data, or an internal trend desk.

scripts/adapters/ · scripts/services/signals.mjs
02 / Judgment

Change the decision contract

Prompts and strict JSON Schemas are repository files, not instructions hidden inside an app.

scripts/prompts/ · merch/weekly/schemas/
03 / Product

Change the physical format

Add another garment, print technique, panel system, or deterministic composition grammar.

merch/base-products.json · scripts/services/weekly-product.mjs
04 / Outcome

Change where it lands

Keep the Preview, connect another provider, or route validated concepts into an existing commerce stack.

scripts/services/production-providers.mjs · app/

Commercial thesis

A small proof for a large fashion problem.

Fashion groups invest heavily in speed, signal quality, and personalization. This project makes that signal-to-product loop visible, modular, and testable at garment level.

Speed

Shorter signal-to-sample loops

Turn a qualified cultural signal into a testable product system in one traceable run—relevant to high-velocity retailers such as Zara and Shein.

Specificity

Smaller, sharper capsules

Translate community, geography, or customer-segment signals into micro-runs without rebuilding the creative and production workflow each time.

R&D

A replaceable fashion lab

Luxury groups such as Richemont and LVMH could evaluate models, sources, product formats, and approval policies as separate components instead of one opaque generator.

Named companies are market examples only. No affiliation, endorsement, customer relationship, or use of their proprietary data is claimed.

Honest boundary

The proof is live. Commerce is gated.

The same creative pipeline can end at a safe concept Preview or continue through provider and storefront adapters. Those are intentionally different permissions.

Now / Visible

Build Week Preview

Generated garments, catalog data, public rights notes, and the system story are available to inspect. No account or API key is required, and checkout is disabled on both client and server.

Browse the proof garments
Now / Live storefront

Production release

Real purchases use Stripe Checkout, verified webhooks, durable order state, and an idempotent Printful draft for manual production review. This is fan-made, unofficial merchandise.

Review the independent-shop terms