Your data stays in the fold.
AI that is safe, powerful and tuned to you. Open models on European infrastructure that answer from your documents and never send a byte outside.
Pick a request
Can the tenant leave the flat early?
- Routing
- Legal reasoning modelA question of statute and contract: it goes to the model that reasons best in law.
- Your documents
- Residential lease 2024 · clause 7Spanish Urban Leases Act (LAU) · art. 11
- Your adapter
- Law firm · internal memo
- Answer
- Yes. After six months the tenant may withdraw with 30 days' notice (LAU art. 11). Your contract also sets compensation of one month's rent per year remaining (clause 7).
- Safe. Everything runs on EU-owned infrastructure. Nothing leaves.
- Powerful. The router picks the best open model for each request.
- Tuned to you. Your documents give the content. Your adapter gives the form.
Illustrative demo with sample documents. Redil opens in 2027.
Coming soon. First access in 2027.
Safe.Your data stays in Europe. And nobody can make us hand it over.
What changes with Redil is who can be compelled to hand over your data, and who does the work.
- Who can be compelled to hand over your data?
- With RedilNo one outside the European Union
- US APIThe provider, under the CLOUD Act and FISA 702, wherever the server isOpen model on your own serversNo one outside your jurisdiction
- Who runs the GPUs?
- With RedilRedil
- US APIThe providerOpen model on your own serversYour team
- A model tuned to your documents?
- With RedilYes, on shared GPUs
- US APIDepends on the providerOpen model on your own serversYes, on dedicated hardware
- Which model answers you?
- With RedilAn open one, always identified
- US APIA closed oneOpen model on your own serversThe one you install
- And the compliance paperwork?
- With RedilIncluded with every deployment
- US APIYours to doOpen model on your own serversYours to do
| US API | Open model on your own servers | Redil | |
|---|---|---|---|
| Who can be compelled to hand over your data? | The provider, under the CLOUD Act and FISA 702, wherever the server is | No one outside your jurisdiction | No one outside the European Union |
| Who runs the GPUs? | The provider | Your team | Redil |
| A model tuned to your documents? | Depends on the provider | Yes, on dedicated hardware | Yes, on shared GPUs |
| Which model answers you? | A closed one | The one you install | An open one, always identified |
| And the compliance paperwork? | Yours to do | Yours to do | Included with every deployment |
This is how the service will work when it opens, in 2027.
- Only open-weight models under permissive licences such as Apache-2.0 or MIT.
- No data or usage deals with the makers of the models.
- Your data, indexes and adapters are yours and train nothing else.
Powerful.Several open models, and the best one for each request.
You do not depend on one model or one company.
Each request, to its best model
A router built on open decision models spreads requests across several models. Reasoning, extraction, speed: each one goes where it is handled best.
You always know which model answers
Only open-weight models under permissive licences such as Apache-2.0 or MIT. No data deals with the people who make them.
Tuned to you.An AI of your own, without buying a single GPU.
Redil turns a general model into yours.
It answers from what you already know
It searches your case files, protocols or regulations and answers with the source cited. Anyone can check it.
It speaks like your organisation
An adapter trained on your data sets your format, your tone and your terminology. It is yours and trains nothing else.
It costs close to a shared model
Thousands of adapters share the same base weights and the same GPUs. Your model no longer needs your own hardware.
For those who hold data that cannot leave.
Law firms
Professional secrecy and client files.
I'm at a law firmHospitals
Patient records, protocols and your own guidelines.
I'm at a healthcare providerPublic bodies
Case files and procedure, with Spain's ENS on the roadmap.
I'm at a public bodyCompanies
Intellectual property, contracts and customer data.
I'm at a company
Coming soon.
Redil is being built. This is what exists today and what comes next.
- Today
We are building the engine that serves thousands of adapters on one base, and the decision router.
- 2027
First access, in order of sign-up and by sector.
- After
General availability. ENS certification is on the first-year roadmap.
Built by Ennio Maldonado, an engineer with five years on data and AI platforms at Silicon Valley technology companies, the last one serving open models in production.
Join the list.
We will write when access opens. Sooner, if you tell us about your case.
Frequently asked questions
When will it be available?
First access is planned for 2027. We will write to the list in order of sign-up and by sector.
Where will my data be?
On EU-owned infrastructure, inside the EU. No company subject to United States or Chinese law will have access to it.
Which models do you serve?
Open-weight models under permissive licences such as Apache-2.0 or MIT. You will always know which one answers, and European models will be available for those who require them.
Some open models come from China. Is that a risk?
The models run only on our infrastructure, so nothing reaches the people who made them. You also decide which models you accept.
Do you hold Spain's ENS certification?
Not yet. It is on the first-year roadmap and we will say so here when we have it.
What will it cost?
There are no public prices yet. The aim is that a model of your own costs close to a shared one.
What do you do with the details from this form?
We keep them encrypted in a database in the EU, only to notify you and reply to you. The email we send you passes through Google Workspace. Every email carries a link that deletes your details at once.