Product
RAGnaroX – the auditable RAG and ChatOps stack
A resource-efficient AI assistant that runs entirely on your hardware: modular data integration, hybrid search and function calls into your existing systems, developed in Rust and operated with compact small language models.
How RAGnaroX works – two building blocks
1. Data integration
RAGnaroX connects your existing sources modularly, such as GitLab, Redmine, file servers, wikis and email. Documents are converted to Markdown, split along their heading hierarchy and stored with semantic and sparse vectors. Every knowledge item retains source and timestamp.
2. Search and answer generation
For each question, RAGnaroX combines keyword search and semantic search, removes duplicates and reranks results. A compact model decides whether functions in your systems should be called via MCP. The answer is streamed and includes the source passages used.
ChatOps: knowledge that acts
RAGnaroX does more than answer questions. It makes retrieved knowledge operational. Through the Model Context Protocol, the assistant calls registered functions in your systems. A support employee can research product documentation and create a new ticket in the same chat, without changing tools.
- GitLab: create issues and retrieve project information
- Redmine: connect tickets and wiki content
- Extensible: any system with an MCP interface can be connected
Modest hardware, measurable performance
The target setup is a commodity computer with 64 GB RAM and a graphics card with 24 GB VRAM, such as an NVIDIA RTX 4090. Models below 5 billion parameters achieve response times below 2.5 seconds for simple questions and below 3.8 seconds for multi-hop questions in our setup.