RAG and agents
Assistants that search your documents and your systems, and agents that carry out tasks over several steps, connected to the tools you already use.
Private AI · self-hosted
We design Generative AI systems that run on your infrastructure and take them into production: models trained on your domain, tailored datasets, RAG pipelines and agents. Behind us are more than thirty years of IT projects for businesses.
A ReDiX Informatica project
Since the 1990s ReDiX Informatica has designed and run the systems its clients work on. ReDiX Labs brings the same method to Generative AI: open models, installed on your infrastructure, with confidentiality that comes from how the system is built rather than from a promise written in a policy.
What we do
A demo wins over a meeting. A system in production has to work every day, on your data and within your constraints: that is the part we take care of.
Assistants that search your documents and your systems, and agents that carry out tasks over several steps, connected to the tools you already use.
Language models trained on your domain, for the cases where a general model gets the terms, the formats or the tone wrong.
Datasets built for a specific task, generated or curated by hand, to train a model, evaluate it or feed a RAG pipeline.
Reports that fill themselves in, and workflows where AI takes on the repetitive part, inside the tools your company already uses.
An assistant that works on your company's documents, not on the web. You put them in a space, ask the way you would ask a colleague, and every answer tells you which document it came from. If the information is not there, it tells you so.


{"vendor": "ACME S.r.l.","invoice_no": "2026-0412","total": 1240.00,"currency": "EUR","confidence": 0.98}
You give us messy documents and we give you back data that is structured, organised and checked against what you asked for. On your infrastructure, through an API.
How we work
A method refined over thirty years of projects: start from open foundations, keep data in-house and deliver systems that hold up under real use.
We start from your data and from how you work today. No ready-made demos: we look at the real problem, the constraints and the points where AI is actually useful.
We work with open-source models installed on your infrastructure. No ties to a vendor, full control over the data and costs that follow actual use.
Whether it is a RAG pipeline or a model trained to measure, we take it all the way to production, with monitoring and evaluation, ready for the people who will use it every day.
An AI system is never finished. We measure it, correct it and grow it together with your team.
Years of ICT projects
Building with GenAI since
Open-source foundations
Open source
Some of our models and datasets are published on Hugging Face: research-ready data, models built for Italian, R&D experiments.
Phi-3.5-mini-instruct trained on our DataForge dataset, a collection of multi-turn conversations in Italian and English.
202,000 question, context and answer triples, all in Italian, built for training RAG systems.
An instruction dataset generated with Llama-3.3-70B, served via regolo.ai.
Let's talk
Tell us what data you have and what you would like to achieve. We will tell you frankly what can be done with AI installed on your infrastructure, and what cannot.
Via Goffredo Mameli, 41/A - 20851 Lissone (MB), Italy