How this website works

This website doubles as our proof of concept: it is run by the same kind of systems we build for clients. Every editorial decision on this website is a Git commit.

The console: search, ask, navigate

The kontrollfeld_ wordmark in the header doubles as an input line: start typing or press ⌘K or / — the console suggests pages, services, case studies, articles and FAQ. The index behind it is built from the site's real content at build time and searched entirely in the browser, without a server request.

In the next phase, free-text questions will be answered via Retrieval-Augmented Generation (RAG) on a self-hosted Typesense instance: Typesense retrieves the passages matching your question from the same content index, and a language model turns them into an answer — with references to its sources. Until then, the console honestly tells you that this stage is still to come whenever a question matches no page.

The content pipeline in short

  1. A weekly, versioned pipeline run scans our work (repos, approved documents) and proposestopics — not yet texts.
  2. A human decides per topic through a file change in the repo:status: approved or dismissed.
  3. Only approved topics become drafts — sanitized in multiple stages (secret scanners, PII detection, pseudonymization), critiqued by a second model and submitted as a merge request for human review.
  4. AI-generated content carries a visible provenance badge and references its sources.

Stack

Astro (SSG) · TypeScript pipeline · GitLab CI/CD · Coolify (self-hosted) · MCP server