For Shopify app developers and studios

A new version of your Shopify app in weeks, not months.

One loop from what merchants say to tested, reviewed code. It is the system we rebuild our own apps with, set up for your team and kept updated.

Request an estimateSee how it works

  • On the Shopify App Store since 2013
  • Three of our own apps redesigned in it
  • Your code stays in your repositories

With AppsBuilder

A new version of an app

Before: MonthsWith AppsBuilder: 5 weeks

Bringing one app up to date

Before: $30–50KWith AppsBuilder: $5–15K

5 weeks is our fastest measured case. The cost is our estimate. How we count

at most, from design to developer tickets on a working app
7 days
quality gates between an approved plan and QA
8
commands, agents and skills in Claude Code
121

Stop losing months in the same six places

We lost them there too. Here is the fix for each.

  • The backlog is ranked by whoever asked last.

    Requests are weighed against your tickets, reviews and analytics.

  • Every hand-off loses context.

    Evidence, decisions, the prototype and the specs live in one repository.

  • AI agents write Shopify code from memory.

    Every change is checked against Shopify's current APIs and rules.

  • Each developer's agent behaves differently.

    One set-up for the whole team, updated for everyone at once.

  • A lesson learned on one app is lost on the next.

    Lessons and bug patches are shared across all your apps.

  • Estimates are guesses.

    Estimates are compared with actual hours on one dashboard.

Request an estimateSee the plans

Three steps from what merchants say to tested code

  1. 01

    What merchants say

    Tickets, reviews, analytics and competitors go into one knowledge base.

  2. 02

    A prototype and tickets

    A clickable prototype on Shopify's components, and specs that become Jira tickets.

  3. 03

    Tested, reviewed code

    One pipeline in Claude Code: tests first, reviews, browser tests, then QA.

/appsbuilder:dev

  1. RED testsnew tests fail for the expected reason
  2. ImplementDONE, with the files changed
  3. GREENlint and tests green
  4. Spec complianceCOMPLIANT
  5. Code reviewAPPROVED
  6. Browser testsPASSED: every P0 and P1 case
  7. Docs refreshproject memory updated in place
  8. Lessonszero or more lessons recorded

Handoff to QA

A sketch of the pipeline, not a recorded run.

One system for your whole team

For product and design

From what merchants say to tickets developers can build.

  • A knowledge base from your tickets, reviews and analytics
  • A clickable prototype on Shopify's components
  • Specs that become Jira tickets

For developers

From an approved plan to a QA handoff, the same way every time.

  • A plan approved before any code
  • Tests first, then the code, then a review
  • Browser tests in the developer's own Chrome

For the team lead

One dashboard for every app and every person.

  • Projects, people and roles
  • Estimated against actual hours
  • Lessons shared across your apps

See the real output, not a demo

Eight real pieces from a working Shopify app and from the kit itself, with the names removed. Open any of them.

Evidence first

Tickets, reviews, analytics and competitors in one folder, by kind of source. An entry file says how far each layer can be trusted. Tickets weigh most, market research least.

The knowledge base

kb/
  kb-README.md
  context - self - <app> - profile.md
  context - competitor - <A…E> - profile.md       5 files
  synthesis - <app> - market research.md
  raw - self - <app> - support tickets 2023…2026.md  4 files
  raw - self - <app> - analytics - …                 3 files
  raw - self - <app> - reviews.md
  raw - competitor - <A…F> - reviews.md              6 files

A reviews app on Shopify, folder listing, Nov 2026, names replaced

Evidence, weighted

Support ticketsx1.0
Three more kinds of source sit between.
Market researchx0.3

Weights as set in the knowledge base of ECM Bulk Image Resizer, July 2026; the working project weighs its evidence task by task.

Decisions with what they rejected

Arguments happen while it is still text, and the record says who decided and what it costs.

The decision log, one entry

## 2026-08-10 — Both discount entry points are shown to every merchant
Context: “Apply discount” on the Plans page and “I have a coupon” on the install gate were built visible to everyone, with the question of who should see them left open. On the gate the question has weight, because the link sits next to the charge itself.
Decision: Both are shown to every merchant, unconditionally. Product owner, 2026-08-10.
Rejected: Showing the entry only to merchants who arrived with a code (by campaign parameter or account flag) — it would hide the field from merchants who got a code from support or a partner, which is most of the ways a code travels. The known cost is accepted: a visible field teaches merchants to look for a discount before paying full price.
Status: active

A reviews app on Shopify, decision log, Aug 2026, verbatim, bold marks removed

Every state, named

The prototype is built from Shopify's own components. If a state is not on its list, it does not exist.

A component's states

/admin/components/install-gate
  Banner arm      the dashboard is dimmed, the banner is the only live element
  Dialog arm      the dashboard renders under a blocking dialog
  Banner arm, back from Plans with a coupon
  Dialog arm, back from Plans with a coupon
  Dialog arm with the intro offer
  Banner arm with the intro offer (hypothesis: not on prod)
  Banner arm, no trial left
  Redirecting

A reviews app on Shopify, Aug 2026, shortened

A check before anyone builds it

An agent with a fresh context checks the work axis by axis. Each gap gets a one-line fix, and parity with production is measured, not eyeballed.

A verdict

/verify · verdict: SHIP-READY for design review
three gaps found and fixed in the same pass, one accepted delta

[x] c. evidence   every research line carries layer, period, coverage;
                  the stale? ticket is no decision's sole ground

gaps → fixes, all applied
  a: reduced motion had no section of its own → a new section with a live frame
  c: a preview was not flagged → flagged as a hypothesis on the feature page
  g: a card 2 px off the shipped one → line box fixed, as shipped

A reviews app on Shopify, autumn 2026, shortened, measurements left out

A ticket developers can build

Each epic is one spec file: rules, acceptance criteria, QA steps. A script creates the Jira tickets, and it stops while the spec holds an open question.

A spec, one ticket

## 1 · Storage and API
<!-- jira: <KEY> -->
Type: Story · Priority: High
Prototype: <preview>/admin/pages/<page>

Acceptance criteria:
- A name saved with capitals is stored and returned in lowercase.
- Renaming to a name that exists returns an error; nothing changes.

QA:
- Save a name with capitals, reload: the item and the list show it in lowercase.

A reviews app on Shopify, autumn 2026, translated, shortened, the feature and its values generalised

Eight gates, in order

Each gate has its own agent and its own exit. A failure goes back to the gate that owns it, and after three review rounds the developer decides.

The development pipeline

#  gate             who                          exit
1  RED tests        test-writer                  new tests fail for the expected reason
2  Implement        implementer, back and front  DONE, with the files changed
3  GREEN            main agent                   lint and tests green
4  Spec compliance  spec-compliance-reviewer     COMPLIANT
5  Code review      code-reviewer, all lenses    APPROVED
6  Browser tests    browser-tests                PASSED: every P0 and P1 case
7  Docs refresh     main agent                   project memory updated in place
8  Lessons          main agent                   zero or more lessons recorded

The kit's pipeline skill, gate table, Oct 2026, shortened

Browser tests in your own Chrome

Cases are written from the plan and run in the developer's own session. On this feature: 17 cases, 15 passed in Chrome, 2 covered by unit tests, none failed.

A browser test report

Browser Test Report — Share patches & lessons with the organization
Mode: FULL, cases BT-001..BT-017, the developer's existing Chrome session

BT-001  Owner shares project lesson        P0  PASS
BT-008  Double-click confirm               P2  PASS
BT-012  No leak to other org               P0  COVERED-BY-PHPUNIT
BT-016  Origin label with special chars    P2  PASS
…
Console: no errors or exceptions on the project view pages during the run.

The kit's own server, test report of Oct 1, 2026, shortened

Patterns from our live apps

Skills taken from the code of our own apps, with every identifier removed. This one keeps API calls working with Shopify's expiring offline tokens.

A pattern skill

# Shopify Expiring Offline Tokens

Keep every Shopify API call working with expiring offline tokens (~1 h access token + ~90 d refresh token) without races between workers.

- One service owns every token write, always under a per-shop distributed lock.
- Inside the lock, re-read the row … another worker may have refreshed already (double-checked refresh).
- Refresh on read: refresh when token_expires_at <= now + 5 min.
- Never copy the token into anything that outlives the request — queued job payloads, caches, websocket auth. Pass the shop id and read the token in the job.

A pattern from the code of two of our upsell apps, Oct 2026, shortened

A new version in 35 and 45 days, measured on our own apps

Counted from the first commit of each new version to its main branch, autumn 2026.

the new ECM Email & Discount Popup
35 days
the new ECM Post Purchase Upsell
45 days
developer made 123 of the 125 commits of the second
1

A reviews app on Shopify

Start of design to hand-off to developers, calendar days

Feature 1
Same day
Feature 2
4–5 days
Feature 3
6 days
Feature 4
7 days, a feasibility check

Calendar days. Feature 4 went to the developers as a feasibility check.

From the app's repository, read November 13, 2026.

Our own apps

The first was SEO Image Optimizer, now listed as ECM Bulk Image Resizer.

  1. Jun 30, 2026Product profiles
  2. Jul 1, 2026Knowledge base
  3. Jul 2, 2026Prototype of the whole app
  4. Jul 14, 2026First task through the full loop
  5. Aug 27, 2026Listing brief

Dates from the app's repository, read November 2026.

Start with one app, or take the whole system in-house

Every plan gets the whole kit and every update. The plans differ in how much we set up for you.

  • Starter

    For a developer or a small team trying it

    • The whole kit, for every developer
    • The team dashboard
    • Every update from our server
    • You connect Claude Code with our quick start

    Request an estimate

  • Recommended

    Pro

    For an app team that wants it on its own data

    • Everything in Starter
    • Your knowledge base, built from your data
    • Your app onboarded, your tracker connected
    • A design repository with a prototype

    Request an estimate

  • Team

    For a studio, or a team with several apps

    • Everything in Pro, for every app
    • Lessons shared across all your apps
    • Patterns from your own code
    • Your first feature built together

    Request an estimate

  • Own it

    For a company that wants it in-house

    • The server, the dashboard and the kit as source
    • Deployed on your infrastructure
    • Your team trained to run it
    • Our updates by subscription, if you want them

    Request an estimate

Not sure which plan fits? Tell us about your apps.

Prices depend on your apps, people and set-up. We send them with the estimate.

Ask which plan fits

Your questions, answered

Does it replace my team?
No. A person makes every call: a developer approves each plan before any code is written. The same people get through the work faster.
Does your server see our code?
No. Commands run in your Claude Code and the code stays in your repositories. Our server stores what your agents save to it: plans, decisions and lessons. Your knowledge base lives in your own repository.
Which stacks and trackers does it work with?
Claude Code is required. Stack skills cover Node and Laravel; the core and Shopify commands work on any stack. Tasks sync from Jira, Plane, Linear, GitHub, GitLab or YouTrack.
Can we run it ourselves?
Yes, on Own it. We hand over the server, the dashboard and the kit as source, and train your team.
Will it get my app through App Review or Built for Shopify?
Nobody can promise that, and we do not. Each change is checked against Shopify's rules before QA.
Where do these numbers come from?
The time is measured on two of our own apps, from the first commit of the new version to its main branch: ECM Email & Discount Popup, August 25 to September 29, 2026 (35 days, the five weeks at the top), and ECM Post Purchase Upsell, September 13 to October 28 (45 days). The cost figures are our estimates of May 2026, not measurements. Our image resizer was the app we built the system on, so its timeline includes building the system and we do not quote it. The other counts were read from the repositories in October and November 2026, and the kit is counted as of October 22, 2026. We have not measured hours yet.

Get a plan and a price for your apps

Tell us about your apps. We answer with the scope and the price.

We answer to this address.

Optional.

For example: the plan, how many apps and people, the stack, a new version or new features, the tracker you use, the data you have.

Thank you. Your request is in.

This form does not ask for your tickets, reviews or Partner exports.

See our apps