AI agents that know your business.
Web chat, WhatsApp, phone, email. AI agents that work on your real data, not on a generic LLM playground. With a RAG pipeline on your content, MCP tooling for your applications and gold-standard eval datasets to keep hallucinations in check. GDPR-compliant, model-vendor agnostic, hosted in Germany.
We eat our own dog food.
We build our tools for ourselves first. Only when they survive our own everyday work do we hand them on — as open source or as a commercial product. Moselwal uses agent-native patterns in its own operations pipeline — from content distribution to internal support workflows.
Three ways to introduce AI in your setup.
Office Agent Suite
For anyone whose biggest lever sits in office day-to-day work: pre-classify the inbox, prepare quotes, make knowledge from SharePoint and Confluence available. Web chat as the standard channel, optional WhatsApp and phone via business-agent-pro.
TYPO3 Business Agent
For platforms running TYPO3 as a CMS: a conversational agent directly in the frontend that knows your content — with access-class routing (Public, Partner, Internal, Admin), MCP tool integration into the backend, and an embeddable chat widget. Runs on our open-source extension business-agent.
AI integration into your existing IT
For existing systems that aren't TYPO3: SAP, MS 365, your own ERP or CRM applications, ALB stacks. We build RAG pipelines on your real data and MCP servers to your applications — without you having to rethink everything.
How we get to a productive agent together
1–2 weeks. Use-case workshop, data-source inventory, risk assessment, channel choice. Outcome: a prioritised use case and an architecture sketch.
4–6 weeks. First RAG pipeline on a real data slice, MCP server connection to the prioritised application, evaluation dataset with gold-standard answers. You see productive answer quality on your content, not in a demo.
4–8 weeks. Extension to further data sources, additional use cases, audience integration via web chat or — if needed — WhatsApp, phone, email via business-agent-pro. Includes audit log and permission model.
Continuously. Model updates run automatically through the evaluation pipeline. Data sources update incrementally. Quarterly use-case review with your team — what runs, what's coming next, what gets switched off.
What we don't do.
We don't build full-autonomy agents that place orders or make legally binding statements without an audit trail. We don't build chatbots as a marketing gadget with three pre-prepared answers. And we don't swap working legacy systems for an AI stack just because AI is currently loud.
If that is exactly what you want, we are the wrong address. If you want to introduce AI as a tool that gives time back to your team and stays auditable, we are the right one.
Brownfield, model-vendor agnostic, and with our own eval framework.
Often used together.
Which use case is worth tackling first?
30 minutes to find out where AI has the largest leverage in your setup: office routine, CMS workflow, or customer touchpoint. Including an honest recommendation if your next step looks more like an internal pilot than a finished product.