Mistral unveils Forge to help enterprises build their own AI models on internal knowledge

Mistral has announced Forge, a system designed for enterprises to build AI models using their proprietary knowledge. While most AI models are trained mainly on public datasets for general-purpose tasks, Forge bridges this gap by allowing organizations to harness their own internal documentation, codebases, structured data, and operational records during training.

By using Forge, models learn enterprise-specific vocabulary, reasoning patterns, and operational constraints, aligning with the unique requirements present in each organization. This approach enables teams to create AI agents that utilize internal terminology and possess an understanding of their workflows and business processes.

Forge also offers flexibility in model architecture, supporting both dense and mixture-of-experts (MoE) designs. This gives organizations the ability to optimize models for performance, cost, and operational demands. Importantly, all models built with Forge remain under enterprise control, allowing organizations to govern model training and evaluation according to their proprietary policies and operational standards.

Additionally, Forge is built for ongoing adaptation rather than one-time training. Organizations can use reinforcement learning pipelines to continuously refine their models, drawing on feedback from internal evaluations and real operational workflows.