MiMagic builds the foundation layer for European AI knowledge systems. Today: Enchilada, a managed knowledge platform for AI agents. Research direction: world-models architecture and privacy-preserving memory. Open-core. EU-hosted. Made in Germany.
We solved intelligence but forgot memory. The most powerful AI reasoning engines today know nothing about the businesses they serve, and the existing memory layer doesn't operate under European jurisdiction. For regulated European industries, that's a structural blocker.
We're building the missing layer: EU-sovereign knowledge infrastructure for AI agents. Hosted in Europe. Built in the open. Designed for organisations that need their operational knowledge under European jurisdiction: by regulation, by contract, or by design.
Today this means Enchilada, a managed Knowledge Platform with multi-engine RAG architecture. Alongside it, we pursue a research direction into world-models-based knowledge representation and privacy-preserving memory architectures, a generation beyond today's vector and graph retrieval. We are in this for the long horizon.
Current RAG systems retrieve stored items. They do not learn, predict, or generalise across knowledge domains. This is adequate for many workloads. It is not adequate for all of them.
We are investigating three problems alongside our production platform. None are shipped. All are active research directions with a multi-year horizon.
Predictive models that learn the generative structure of knowledge domains rather than storing facts. The hypothesis: a system trained to predict how concepts relate, evolve, and constrain each other can generalise to questions no stored document addresses.
Knowledge representations that permit learned reasoning without exposing tenant data. Current approaches protect vectors but not the computation over them. We are investigating architectures where privacy operates at the reasoning layer, not just the storage layer.
Knowledge systems that improve with use rather than merely accumulate data. The boundary between retrieval and learning is the engineering problem we keep returning to.
A managed Knowledge Platform for AI agents. Knowledge Bases, Skills, and Agents, bundled per domain. EU-hosted. Live, accepting customers.
Visit getenchilada.com →The foundational components of our knowledge infrastructure are open-source. Engine adapters, ingestion connectors, MCP server implementation: designed to be operated independently if external assumptions change.
Commercial value lives in the managed experience and curated knowledge packs. We believe open core is the right architecture for infrastructure-grade software.
Public repositories will follow as the foundational components stabilize.
EU-hosted. Your data stays in European jurisdiction. Zero extraterritorial exposure.
Multi-LLM architecture with European-first defaults: Mistral as primary EU LLM provider. Other providers configurable per workspace, never required.
We commit to open protocols (MCP, OpenAI-compatible APIs) and an open-core licensing posture so the system can keep operating if external assumptions change.
Partnerships, press, general inquiries
hello@mimagic.aiCommercial questions about Enchilada
getenchilada.com → Contact →Entity
MiMagic GmbH