Skip to content
aleph labs

AI that runs on your machines,not on someone else’s.

I install open models on your own hardware, connect them to your documents, and you keep control of what runs.

What I do

  • Search inside your documents

    Search that answers and cites the source, instead of guessing.

  • Agents and automation

    The work that comes back every week, handed to agents that leave a record.

  • Models installed on your side

    An open model on your own server. No data leaves the network.

  • Audit and explainability

    What the system decides, what it relies on, and where it fails.

  • Search visibility

    Pages that go after queries people actually type. Written, reviewed, measured.

  • Google Ads

    Campaigns that run and correct themselves. You read the report.

  • Training

    Half a day with your team: what these models do, and what they don’t.

Selected work

What is running today, at clients’ or at home.

Terminal: two RTX 3090s loaded, the model service active

An AI companion running at home

A 27-billion-parameter open model on two graphics cards. Nothing leaves the machine.

Chart: organic clicks per week over twelve weeks

Search visibility, measured

No page ships without a verified search volume. On vidyalaya.fr: 690 organic clicks in 28 days.

Chart: clicks per campaign over thirty days

Google Ads on autopilot

Four campaigns, 9,823 clicks in 30 days. The account watches and corrects itself.

Terminal: 770,718 passages indexed, none awaiting embedding

Hybrid search over 770,000 passages

Vectors and keywords fused. Every answer cites its source in the corpus.

Terminal: three messaging gateways up for weeks

Three agents on a messaging app

A companion, an artist manager, an operations assistant. They watch and they report back.

The ANTIK online shop

ANTIK’s editorial pipeline

A model writes, a panel of models reviews, nothing ships without a measured search volume.

maisonantik.com
A story published on latence.org

latence.org — a site that writes itself

One story a day: the agent writes it, commits it and deploys it.

latence.org
Home page of the MENTOR app

MENTOR — working out which benefits you qualify for

The eligibility engine, designed and wired in. The app is in production.

mentor-app.fr
Sign-in screen of the Vidyia tutor

Vidyia — a tutor for school students

A teaching assistant plugged into the digital workspace schools already use.

vidyia.com
Fonos: a work declaration form and the sign-in screen

Fonos — music royalties in one interface

Filing works and reading royalty statements in the same place.

fonos.fr
Home page of the Vidyalaya association

Vidyalaya — a charity’s resources

Site, learning portal, and worked answers checked before they are published.

vidyalaya.fr
The CRM sign-in screen, dressed as a vintage television set

The CRM of a video equipment rental company

Clients, quotes, operations schedule, and a read-only sync from their Google Sheets.

crystalguard.fr
Terminal: the ingest endpoint answering

A wearable memory

A small device captures sound and images and sends them back to a machine at home.

Let’s talk.

Describe what you need in a few lines. Answer within two working days.

contact@alephlabs.fr

The form prepares the message in your own email client. Nothing is sent from this site: it has no server, no database and no cookies.