Bringing DevOps & AI to Shopware
Accelerate as a team without losing control, readability or trust. An internal dev-to-prod platform, a quality net at every level and an agentic framework governed by the team.
Pragmatism before complexity
Our stance: keep a robust, understandable stack that the whole team can operate easily. It's this readable, team-governed base that lets us introduce AI without losing control.
- A team of seniors — 10 developers who became seniors in-house.
- Docker from dev to prod — the same stack since 2015.
- Dedicated hosting — on our servers or the client's.
- High resilience — not high availability at any cost, but real resilience.
If our internal dev platform is in PHP, then the whole team is DevOps.
An internal development platform
Built on Castor, a modern PHP task runner: a clean CLI contract at the front, testable PHP business logic at the core. Two engines behind it: a declarative PHP model and GitLab as the execution engine.
Thin tasks, testable core
Castor tasks validate input then delegate. Real logic lives in testable PHP services — immutable Command Builders (Docker, GitLab, SSH) and Command Runners (CI, deployment).
Project › Applications › Instances
A .fops.php file describes everything: applications, extensions, instances and profiles. Multi-repo workspace, services grouped or split by client context, same model locally, in CI and in production.
CI, registries, environments
GitLab capabilities are driven and leveraged: CI, Container & Package Registry, Environments and a rollback button. Multi-server deployment, from dev to prod.
The generic base, specialised for Shopware
Two layers that stack without duplication. FroggShop automatically wires and configures all the official Shopware tooling into the stack, in the right place. Same stack from dev to prod, with Shopware's best practices for performance and production.
Dev & tests wired in
Admin + storefront build, watch / hot-reload, the official Acceptance Test Suite and Admin MCP for the coding agent — wired automatically.
Production best practices
Workers & scheduled tasks, OpenSearch, Varnish with xkey soft-purge, Valkey cache and sessions — configured at the right time.
Official tooling
shopware-cli, shopware-deployment-helper and @shopware-ag/admin-mcp provisioned automatically. No tinkering: we leverage and extend the official tools.
The same quality net at every level
A shared suite across Shopware and every plugin, applied at each level and executed automatically.
Config per level
Shopware and each plugin (FroggShop, Cawl, AlfredSync…) declare their own phpstan.neon, phpunit.xml and .php-cs-fixer.
Pre-commit
Staged files are checked extension by extension, installed automatically in dev.
Auto-generated CI
Independent jobs per extension, generated from the PHP model and run in parallel.
Harness engineering, as a team
A coding agent is an ultra-experienced senior developer… but it's their first day on the project, every single task. The project harness completes the agent's harness: it opens the existing base to agents, without creating a second tooling chain.
Documentation that executes
AGENTS.md routes and maps, ARCHITECTURE.md sets boundaries, PLANS.md enforces a plan. Doc quality is checked like code: "what doesn't live in the repo doesn't exist".
The right context, not all of it
Targeted context, fed by real runtime introspection. On storefront search, we go from a global context of about 47,500 tokens to a focused one of about 6,700 — nearly 7× less.
We decide upfront what must be proven
The validation contract is defined by the project harness, not the agent. Building on the Shopware Acceptance Test Suite, we replay real journeys and capture evidence at key moments, tied to the business.
A prompt is hope, a script is knowledge.
Team in the loop, not the other way around
AI doesn't write in our place: it shifts the work towards design and decision. The team specifies, the dev plans, the agent executes, the team validates. A loop, not a straight line. It's not a lone dev reviewing the agent: the team frames, validates and maintains.
Let's be honest: we're learning
Models improve fast but also regress, pricing shifts, and the project harness lives like code: it must be maintained. Our fixed point isn't the tool of the moment, but the team's ability to learn fast, verify what it adopts and drop the rest.
Want to talk DevOps, AI and Shopware?
Let's discuss how to industrialise your Shopware projects without losing control.
Contact us