TildeLM: foundational LLM for a multilingual Europe

TildeLM

The base layer for your next AI product

TildeLM is a foundational large language model. It means that it’s a powerful multilingual base you can fine-tune and build on. Designed for flexibility, it can be adapted to power a wide range of AI solutions tailored to your specific needs.

USE CASES BY INDUSTRY

One model, endless possibilities

For governments 🏛️

Governments can use TildeLM to build national language models that ensure fair access to public services, support digital sovereignty, and strengthen linguistic inclusion across all official languages.

For businesses 💼

Your business is unique and so should the AI solutions you use. TildeLM lets companies build secure, customisable language models that understand your specific terminology, context, and audience.

For research & innovation 💡

TildeLM supports researchers in studying languages, improving translation systems, and developing innovative language technologies. It provides a flexible base for exploring new ideas and enhancing language tech capabilities.

For healthcare 🏥

Healthcare providers can leverage TildeLM to build AI systems that accurately process multilingual medical data, support cross-language patient communication, and provide high-quality multilingual health information in various languages.

For financial services & insurance 📈

Financial institutions and insurance companies can use TildeLM to improve customer service, automate document processing, and handle multilingual financial data securely.

As an open-source model, TildeLM will be yours to use, adapt, and fine-tune – whether you’re driving research or creating custom AI tools.

Fine-tuning LLMs for enterprises with FFplus

With support from the FFplus initiative, we’re fine-tuning TildeLM and other LLMs for a secure, locally deployable enterprise search and Q&A system. This project focuses on smaller European languages, helping organisations use AI safely – without sending data to third parties – and ensuring better quality than generic English-centric models.