Sovereign AI for PDF Intelligence – Multimodal, Local, Efficient

Organisation Overview

What is Elixir?

Elixir develops sovereign multimodal AI models that extract, structure, and activate data from complex PDF documents—locally, without external dependencies, and under your complete control. We advance this work in partnership with the Intelligence Lab at ECE, France’s first Gen AI research hub.

We specialise in regulatory and sensitive contexts across finance, legal, and the public sector, beginning with recurring documents such as Key Information Documents (KIDs), financial reports, and technical annexes.

Our Models — Small, Specialised, Sovereign

We design and fine‑tune our own compact language and vision–language models—collectively named SAGE (Sovereign AI for Governance & Extraction)—optimised for the realities of regulated organisations.

Key characteristics

• Runs efficiently on a standard CPU or Apple silicon
• Purpose‑built for heterogeneous, real‑world document layouts
• Trained exclusively on proprietary, in‑house datasets (see Elixir Corpus)

A selection of these models is available for open evaluation on Hugging Face Spaces.

Elixir Corpus — A Purpose‑Built Data Foundation

All models are trained on the Elixir Corpus, a structured collection of public and regulatory documents.

Each subset targets a specific domain—finance, public governance, legal frameworks, ESG reporting, and more—and includes:

• Text‑based and scanned PDFs
• Tables, text, images, and charts