Aleph Alpha

Aleph Alpha builds customised large language models it calls sovereign LLMs, for organisations that cannot route their data through a general-purpose model API. Rather than selling one shared endpoint, the company designs and trains a model against a customer's own domain knowledge and then runs it on European infrastructure under EU law. Delivery goes through PhariaAI, an inference and application stack shipped as Kubernetes Helm charts that can be installed on the customer's own hardware, in a private cloud, or in an air-gapped network.

Customers receive the full LLM checkpoint, weights and code, for commercial use alongside the runtime; default model sets include luminous-base, Llama 3.1 8B Instruct, Llama 3.3 70B Instruct and Llama Guard 3, with PostgreSQL and RabbitMQ used for persistence and queues. Pharia-1-LLM-7B weights are also published on Hugging Face under the Open Aleph Licence, which limits use to education and research. References published by the company include an AI assistant for a public authority serving 80,000 users, a global chip manufacturer that cut document search times by 90%, and an automotive supplier that processed RFQs 40% faster.

Aleph Alpha was founded in Heidelberg in 2019 and is still headquartered there, developing from four German locations with an international team of engineers and scientists. It places itself in industry, the public sector and defence, and its leadership is organised around co-CEOs and co-chief research officers.