On-premises AI for SMEs
Your company knowledge. Your hardware. Your AI.
RAGnaroX is an AI assistant that answers questions directly from your own documents, tickets and wikis, and can act on request. Fully operated locally, without cloud dependency, without US providers and without data leakage.
100%
On-premises: no data leaves your network
0.90
Context precision for single-hop questions (SQuAD benchmark)
2.5 s
average response time per request
ICSE 2026
scientifically published and independently reviewed
The problem
Cloud AI creates dependency, and your data travels with it
Using generative AI today almost always means using external providers. Data leaves the company, request costs rise, and in regulated sectors such as finance or healthcare auditability becomes a compliance risk. At the same time, your organization’s real knowledge is scattered across file servers, wikis, ticket systems and email, making it hard to find in daily work.
The answer
RAGnaroX: find instead of search, on your own hardware
RAGnaroX combines retrieval-augmented generation with ChatOps. Before every answer, the language model checks your own source material. Via the Model Context Protocol (MCP), the assistant can also perform actions in your systems, for example creating a ticket in GitLab or Redmine. The complete stack is developed in Rust, runs on commodity hardware and remains fully under your control.
Three good reasons for locally hosted AI
Data sovereignty
All models run locally via llama.cpp. Documents, questions and answers never leave your network. GDPR compliance becomes an architectural property, not a contractual promise.
Predictable costs
No per-token fees and no vendor lock-in. RAGnaroX runs on a commodity machine with 64 GB RAM and a 24 GB graphics card, with requirements decreasing as compact language models improve.
Auditability
An open Rust-based stack instead of a black box. Every answer includes the source passages used, making results traceable for regulated industries and the EU AI Act.
What RAGnaroX handles in everyday work
Find knowledge
Documentation, wiki pages, issues and files become centrally searchable, with precise answers and source references instead of result lists.
Execute actions
Via ChatOps, the assistant creates tickets or calls functions in your systems directly from the chat, connected through the open MCP standard.
Work multilingual
German, English and Spanish were tested as part of the scientific evaluation, including cross-language search across mixed data sets.
Ready for AI without cloud compromises?
We will show you RAGnaroX live, optionally with an excerpt of your own data. Non-binding, in 30 minutes, remotely or at your location.