RAG Enterprise: Query Your Documents with AI Without Sending Them to the Cloud

Imagine being able to ask a complex question across all your organization's documents — contracts, manuals, archived emails, reports — and receive a precise answer with source citations. This isn't science fiction: it's what RAG (Retrieval-Augmented Generation) does. The problem is that almost every available solution requires sending those documents to American servers. I3K RAG Enterprise starts from the opposite assumption.

The enterprise AI paradox

The organizations that would most benefit from AI to manage large document volumes are often those that can't use it: law firms with confidential dossiers, healthcare facilities with medical records, public offices with sensitive citizen data, companies with trade secrets. SaaS solutions assume documents go to the cloud. I3K RAG Enterprise assumes exactly the opposite.

The platform comes in two versions with very different capabilities — and it's important not to confuse them.

Community edition: open source on GitHub

The Community version is released under the AGPL-3.0 license and freely available on GitHub. It installs with a single command on Ubuntu 20.04 or later and can operate completely air-gapped — isolated from the network, with no calls to third-party APIs.

It includes the complete RAG pipeline, web interface, APIs, multilingual support across 29 languages, and the backup system. It's the starting point for organizations that want to evaluate the solution or have an internal technical team capable of managing installation and maintenance.

Pro edition: the features that make the difference

The Pro version, available at rag-enterprise.com, adds capabilities that Community doesn't have — and that for many enterprise organizations are essential.

The most significant is SQL extraction: the system can query structured databases using natural language, not just documents. You can ask "how many invoices over €10,000 were issued in Q1" and get the answer without writing a single line of SQL. This capability does not exist in the open source version.

Pro also adds enterprise SSO (Active Directory, SAML, OIDC integration), advanced audit logs that meet enterprise compliance requirements, and dedicated support SLAs.

How the RAG pipeline works

Regardless of edition, the RAG pipeline runs in four phases:

Ingest: you upload documents via web interface or API. The system uses Apache Tika and Tesseract to extract text from PDF, DOCX, PPTX, XLSX, ODT, RTF, HTML, XML, and scanned documents via OCR.

Embed & store: documents are split into semantic chunks and transformed into vectors using BAAI/bge-m3 (29 languages, no per-language fine-tuning). Vectors are stored in Qdrant with metadata for RBAC filtering.

Retrieve: your question is compared semantically against the vectors. This isn't keyword search — the system understands meaning. Relevance threshold, top-K, and role-based filtering are all configurable.

Generate: relevant chunks are passed to the LLM — by default EuLLM with Qwen3:14b or Mistral 7B Q4 — which generates a response anchored to the documents. Everything locally, zero external calls.

Hardware, backup, and compliance

The platform runs on hardware you already own: NVIDIA CUDA, AMD ROCm, or CPU-only. Each node handles over 10,000 documents. The integrated rclone backup supports over 70 providers (S3, MEGA, Google Drive, OneDrive, Dropbox, Nextcloud, Backblaze B2, and others) with cron scheduling and zero-downtime.

JWT authentication, three-role RBAC, and audit logs ensure the traceability required by GDPR and the EU AI Act. The Community version source code is auditable end-to-end.

Who needs it

If your organization needs to query confidential documents with AI and cannot send them to American clouds, RAG Enterprise is likely the most complete answer available in Europe today. To start with the open source version: github.com/I3K-IT/RAG-Enterprise. For enterprise features including SQL extraction: rag-enterprise.com.