Sovereign AI, on the computers you already own

Cydral Technology is a French open-source initiative. Its C++ library and its two free Windows applications, Albireo and Pseudalis, run language models and protect personal data directly on your own machine: no account, no subscription, no cloud.

Chart of the constellation Cygnus19h3020h20h3021h+30°+40°+50°Denebα CygSadrγ CygAljanahε CygFawarisδ CygηζικAlbireoβ Cyg
Local by design
Your models, documents and keys stay on the computer. The products work without a connection.
Free and open source
The library and the Community editions are published under the Boost Software License 1.0, for personal and professional use.
Built for laptops
Quantized models and optimised kernels: the default assistant runs with 8 GB of memory.
Measured in public
Every figure comes with its hardware, its model and its protocol, in reports anyone can check.

Two products, one library

Everything is written in C++, from the tensor operations to the Windows installer. The library provides the engine; the products bring it to every desktop.

Measured results

Each figure comes from a public report or README of the repository and states the conditions of its measurement.

SubjectResultConditions
GPU memory extended in softwareA working set of 17,506 MiB trained on a card of 12,288 MiBModel of 270 million parameters, batch of 8, budget of 8,192 MiB
Personal data detectionF1 of 0.946 in five languages; 2 of 1,223 values left readable on sixteen test filesPseudalis 1.0.11: synthetic test set in five languages, exact spans, every category; Word, PowerPoint, PDF, CSV, source and image files written for the test
Generation on a laptop GPU56.8 tokens per second, against 23.6 on the processorQwen3-VL 4B in Q4_K_M, GeForce RTX 5070 Ti Laptop GPU, 24 processor threads
Lossless text compressionenwik9 in 117,169,624 bytes, 0.937 bits per byteDecompressor of 7,063,888 bytes, tokenizer and network included
Training routed to the integrated GPU63.3 ms per step instead of 97.4 msIntel Core Ultra 9 275HX, automatic routing, final loss within 0.001 of the host's
Reading a prompt on a laptop216 tokens per second with Qwen3-VL 4B, more than 1,300 with Qwen3 0.6BIntel Core Ultra 9 275HX, 24 threads, processor engine with its AVX-VNNI kernels, first pass of a prompt of 512 tokens

All results and reports

What drives us

Sovereignty

Models, documents and keys stay on machines their owners control. The code can be read, audited, modified and self-hosted.

Frugality

Small models, quantized weights and careful kernels let a laptop do work usually sent to a server.

Open research

Methods, measurements and limits are published with the code, in reports written to be checked and reproduced.

Read our commitments

Developed in the open

Our public repositories cover transformers and language models, tokenization, image colorization, generative adversarial networks and self-supervised vision.

Black and white pictures and their versions colorized by the high-resolution model of Colorify

Join the initiative

Investors, distributors and sponsors: fund the roadmap, distribute the products under your own brand, or add them to your offering.