AI into your existing IT — not instead of it.
Brownfield AI integration for the mid-market and SMEs: RAG pipelines, MCP servers, clean connectors to your TYPO3, Symfony, SAP, ERP, CRM. No AI startup pitch, no greenfield rebuild — AI that plugs in exactly where your data already lives.
Your data lives in your stack — not in an AI demo.
The solution
- RAG pipelines that tap into your real data sources — Confluence, TYPO3, SharePoint, databases
- MCP servers that let AI agents write into your applications under control (e.g. the TYPO3 backend)
- Data classification and policies before any integration — not after
- Evaluation pipeline with gold-standard datasets that run alongside every model update
- Multiple model providers in parallel — no lock-in, cleanly swappable
- A clean connector layer to your TYPO3, Symfony, SAP, ERP, CRM
The problem
- AI tools assume all your data is already in some cloud — it isn't
- RAG demos built in 5 minutes, not scalable in production
- No data classification, no access policies, no auditability
- Hallucinations because no evaluation pipeline exists
- Lock-in with the AI vendor, no plan B
- Connecting to existing systems sold as “integration is the customer's problem”
Four building blocks that get AI to actually reach your stack
These four components are the scaffolding of every AI integration we do — regardless of whether the end goal is an internal assistant, a customer-facing agent or a content pipeline.
Evaluation pipeline
For each use case, a gold-standard dataset with expected answers. With every model swap or prompt update the evaluation runs automatically — with a clear pass/fail threshold. So you spot hallucinations before your users do.
Connectors to your existing systems
Clean integration with SAP, MS 365, ERP, CRM, PIM, DAM systems — either via existing APIs or via MCP wrapping. None of the “well, SAP first has to build a REST API”.
MCP servers to your applications
We build MCP servers that let AI agents access your applications under control — e.g. the TYPO3 backend, Sylius shop, internal tools. With permissions, audit log and clean tool definitions. Exactly the building blocks we also use in AI-Ready CMS as a Service.
RAG pipelines on your real data
Indexing of your real sources (TYPO3, Confluence, SharePoint, files, databases). Chunking, embedding, hybrid retrieval, re-ranking — not from a demo tutorial, but production-ready. Including a clean update strategy when the source changes.
How AI lands in practice
Four posts from the blog on concrete AI integrations — content migration, support filtering, agent readiness, editorial use.
Level 5 Agent-Native: 100/100 in the Cloudflare check
What it means for an existing platform to become agent-native — the structural precondition for AI integration to be sustainable.
AI pre-filters support requests
How we use RAG and MCP to bring context from Git, monitoring and docs into tickets — in a running support setup.
When AI feeds on your content
What it means when AI agents reach directly into the content — and which architectural preconditions that imposes.
Content migration used to be a team. Today it's an agent.
How we run migrations with AI agents — a concrete example of brownfield AI integration.
Brownfield. Model-vendor agnostic. With an evaluation pipeline. GDPR-compliant.
Often used together.
Which data source, which use case, which order?
In 30 minutes we'll work out which use case is worth tackling first for AI in your existing IT — with a rough estimate of effort and risk. Free of charge, no obligation, no sales pressure.
Oder direkt schreiben: kontakt@moselwal.de