Pleias is a French AI lab with its office at Station F in Paris. It develops small language models, such as the Baguettotron models, and open datasets, including Common Corpus and SYNTH. The models are aimed at tasks such as document processing and retrieval-augmented generation. Models and datasets are published on Hugging Face; for organisations Pleias also offers solutions that can run on-premises, such as Synth and Stratum. According to its website, Sillon, a model built for RATP, runs on that organisation's sovereign infrastructure at Scaleway.
Pleias is based in France; the website names no legal entity or owner. The Baguettotron model card states the Apache 2.0 licence; for other models and datasets the Hugging Face page shows no licence. Because weights and data can be downloaded, organisations can host the models themselves and keep data inside their own environment. The website gives no information about the hosting of Pleias's own services. The site links to a GDPR notice.