Sovereign RAG

What is RAG?

RAG (Retrieval-Augmented Generation) is an architecture that enriches a language model’s (LLM) responses with real documents, dynamically extracted from your knowledge base. Rather than relying solely on the model’s training data, the RAG pipeline retrieves the most relevant passages from your documents and feeds them to the model to generate responses grounded in your data.

This is the technology at the core of Lexiane — and it runs entirely on your infrastructure.

What sets Lexiane apart

Most RAG solutions rely on cloud APIs: your documents are sent to third-party services for indexing, search, and processing. Lexiane works differently.

The Lexiane engine is a self-contained Rust binary. Embedding and inference models (Candle, Mistral.rs) are built directly into it. Once deployed, no external connection is required — not even for model inference. Your documents never leave your perimeter.

The pipeline at a glance

Step	What happens
Ingestion	Your documents are loaded, split into chunks, and cleaned
Indexing	Chunks are converted into vectors and stored in the local vector store
Retrieval	On each query, the most relevant passages are retrieved via semantic search
Generation	Context is passed to the LLM, which produces a response grounded in your data

Every step runs on your infrastructure. Every operation is logged in a SHA-256 cryptographic audit chain.

Deployment modes

• Sovereign (air-gapped) — no outbound connections, compatible with classified and isolated networks
• Hybrid — on-premise core infrastructure with selective external connectivity
• Cloud — deployment on contractually defined cloud infrastructure

Articles in this section

• Introduction to RAG — detailed walkthrough of the Lexiane pipeline
• Hexagonal architecture — why Lexiane is built to withstand technological change