i3k AI Software Company Milano - hero background
Italian deep-tech studio · Founded 2015 · Based in Milan

EU-sovereign AI infrastructure for regulated industries

We build AI tools designed in Europe from scratch, with compliance built-in. For organizations that can't put their data in US clouds.

Apache 2.0 · AGPL-3.0
EIC Accelerator 2026 applicant
45+ stars on RAG Enterprise
Founded 2015 · Built in Milan
Scroll

Two complementary products, one thesis

RAG Enterprise for on-premise document intelligence. Powered by EuLLM Engine — our EU-sovereign runtime built on llama.cpp and TurboQuant. Two products, one infrastructure, one thesis: give regulated European organizations the tools to replace American AI SaaS without compromises.

RAG Enterprise

On-premise document intelligence. Open core: AGPL-3.0 + Pro 2.0.

  • Hybrid SQL-Vector
  • Anti-hallucination
  • Air-gap ready

EuLLM

Vertical AI foundry. From 70B generalists to 7B EU-domain specialists.

  • Rust engine
  • Distillation pipeline (Forge)
  • EU-hosted registry (Hub)
AI Enterprise Solution

RAG Enterprise

A 100% on-premise Retrieval-Augmented Generation system. Your documents never leave your server.

Open source community: 45+ GitHub stars · Active development

Open Source

Community Edition

AGPL-3.0 License

  • Full RAG — upload, query and answer from documents
  • Multi-user authentication with JWT
  • Upload PDF, DOCX, TXT with OCR (Tika + PyMuPDF)
  • Vector Search with Qdrant
  • Local LLM via Ollama (Llama, Mistral, etc.)
  • Docker Compose — deploy in minutes
  • User management with admin panel
  • Base React UI included
Enterprise

Enterprise Pro 2.0

Proprietary License

  • Everything in Community, plus:
  • Cross-Encoder Re-ranking for more accurate answers
  • Hybrid Search — BM25 + Vector + RRF combined
  • Multi-Query Retrieval for complex queries
  • Built-in Analytics Dashboard
  • Persistent Chat History (SQLite)
  • Modern vertical Chat UI with message copy
  • Multilingual Reranker (BGE-M3)
  • Full Offline Mode
European Sovereign LLMs

EULLM

Open Source

Open-source platform for European sovereign LLMs. High-performance Rust inference, OpenAI-compatible API, EU-sovereign deployment.

EULLM is the open-source infrastructure for running language models in Europe without cloud dependencies. Rust-based Engine for high-performance inference, Forge for model specialization, and Hub as an EU-compliant registry.

Engine

Engine

EU-Sovereign Inference Runtime

  • EU-sovereign inference runtime — Rust orchestration layer over llama.cpp (MIT, EU-developed)
  • Includes TurboQuant integration (3-bit KV cache compression, ~5x memory reduction, ICLR 2026 algorithm)
  • OpenAI-compatible API. Deployment EU-only, zero non-EU telemetry
  • Streaming, batching and concurrency management
Forge

Forge

Model Specialization

  • LoRA/QLoRA pipeline for efficient fine-tuning
  • Automatic conversion to optimized GGUF
  • Dataset builder from enterprise documents
  • Automated benchmarking and validation
Hub

Hub

EU-Compliant Registry

  • Model catalog with EU AI Act cards
  • Full versioning and traceability
  • One-click deploy to local Engine
  • Verifiable licenses and data provenance

EU AI Act Ready

Model cards, training data traceability and risk classification natively built-in.

Other I3K Projects

I3K Technologies also operates two adjacent product lines, born from earlier R&D efforts. These projects run independently from our core deep-tech focus.

About I3K Technologies

Italian deep-tech studio founded in 2015. Based in Milan. We build EU-sovereign AI infrastructure for organizations in regulated industries that can't — or won't — trust their data to US clouds.

Two core products: RAG Enterprise (on-premise document intelligence) and EuLLM (open source vertical AI foundry). Two adjacent products: CRM81 and LetsAI. EIC Accelerator 2026 applicant.

Founder & CEO: Francesco Marchetti — 27+ years in European IT infrastructure and telecommunications.

How we build

We don't claim to have rebuilt LLM inference from scratch. We build on solid open source foundations like llama.cpp (MIT, developed in Bulgaria) and TurboQuant (algorithm published by Google Research, ICLR 2026, community MIT implementation). We add the missing layer: EU-sovereign orchestration, distillation pipeline, AI Act compliance registry, vertical models trained on European data. The value is not in reinventing primitives — it's in providing the compliance and verticalization layer still missing in the AI ecosystem.

Open source first

Most of our work is open source. Code is the contract.

Compliance by design

AI Act readiness, GDPR, on-premise. Built in, not bolted on.

EU-data residency

Your data never leaves the EU. No exceptions.

Ready to Transform Your Business?

Let's talk about your project. Free initial consultation, no strings attached.

Schedule a Call

Book a 30-minute call

Book now

WhatsApp

Message us on WhatsApp

Start chat

Office

Milan, Italy