Introducing Forge

Mistral AI introduced Forge, a system that lets enterprises build AI models grounded in their own proprietary knowledge, according to the company's announcement.

Forge is designed to train models on internal documentation, codebases, structured data and operational records. Mistral AI said the system supports pre-training to build domain-aware models from large internal datasets, post-training to refine behavior for specific tasks and environments, and reinforcement learning to align models and agents with internal policies, evaluation criteria and operational objectives.

Mistral AI said it has partnered with organizations including ASML, DSO National Laboratories Singapore, Ericsson, the European Space Agency, Home Team Science and Technology Agency (HTX) Singapore, and Reply to train models on their proprietary data.

Forge supports both dense and mixture-of-experts architectures, and multimodal inputs where required, the company said. Mistral AI stated that Forge was built agent-first, citing an autonomous agent such as Mistral Vibe that can use it to fine-tune models, find optimal hyperparameters, schedule jobs and generate synthetic data, while Forge monitors metrics for benchmark regression. The company said Forge handles infrastructure and includes recipes for data pipelines and Mistral AI's own training methods, allowing customization by writing plain English.

According to Mistral AI, Forge allows enterprises to keep models under their control, trained on proprietary datasets and governed by internal policies. The company described evaluation frameworks for testing models against internal benchmarks and compliance rules before production deployment, and reinforcement learning pipelines for continuous adaptation.

The announcement listed potential enterprise applications in government, financial institutions, software teams, manufacturing and large enterprises. Mistral AI invited organizations to sign up to learn more about Forge.