Grubel

Overview

Grubel is a German applied-AI research lab focused on matter-specific AI for complex legal work. Its approach automates a specialization loop that curates relevant data, adapts models and agents to an individual matter, and evaluates results continuously rather than applying one general-purpose system uniformly across cases.

The company operates from Munich and Tübingen and emphasizes data ownership, jurisdiction-specific processing, and local deployment. It is notable for combining research in test-time training and data-centric machine learning with an early commercial focus on legal workflows.

History

Grubel was founded in 2026 by machine-learning researchers Moritz Hardt and Reinhard Heckel. Hardt is a Director at the Max Planck Institute for Intelligent Systems and co-inventor of test-time training; Heckel is a professor at the Technical University of Munich and a co-creator of DataComp and OpenThoughts. The company developed an automated specialization loop combining data curation, test-time adaptation, and continual evaluation for individual legal matters. In September 2026, Grubel announced a €3 million pre-seed funding round led by Point Nine.

Product Lines

Manufacturing

Grubel does not manufacture physical products. It develops software and AI systems, designed for local deployment on customer infrastructure or within the applicable jurisdiction, with the Grubel GmbH operating from Germany and owned by Grubel, Inc. in Delaware.

Notable Products

• Matter-specific AI platform - An AI specialization system that discovers and curates matter-specific data, adapts models and agents at test time, and evaluates outputs continuously.
• Data engine - The data-curation component that identifies task-relevant information and feedback signals for a particular legal matter.
• Matter adaptation - The adaptation layer that specializes a model, agent, and execution harness to a matter before or during task execution.
• Continual evaluation - A continual benchmarking and evaluation component that measures outputs against standards derived from the matter.

Reputation

Grubel is presented by researchers and early-stage technology observers as a research-led entrant addressing the limits of general-purpose AI in complex legal work. Its perceived strengths are the founders’ academic backgrounds, emphasis on evaluation, and focus on matter-specific adaptation and data control. Public evidence remains limited because the company was founded in 2026 and has a small team, while product maturity, customer adoption, pricing, and independent performance validation are not yet broadly documented.