Mistral just launched Forge — a platform to train AI on your company's private data
Mistral Forge lets enterprises build custom AI models trained on their own documents, code, and policies. Partners include ASML and the European Space Agency.
French AI company Mistral just launched Forge, a platform that lets organizations train their own AI models using internal company data — engineering documents, compliance policies, codebases, and years of institutional knowledge.
The announcement, made on March 17, already reached #1 on Hacker News with over 530 upvotes.
Why companies want their own AI models
General-purpose AI models like ChatGPT or Claude are trained on public internet data. They're great at general tasks, but they don't know your company's specific processes, policies, or products. If you ask a generic AI to draft a compliance document for your company, it'll give you a generic answer.
Forge solves this by letting organizations train models on their own data, so the AI understands internal context — specific engineering standards, regulatory requirements, or customer service procedures unique to that company.
Three types of training
Pre-training — Feed the model your company's documents so it understands your domain (like teaching it your industry's language)
Post-training — Refine its behavior for specific tasks (like making it follow your formatting rules)
Reinforcement learning — Continuously improve the model based on feedback (like an employee who gets better with each performance review)
Who's already using it
Mistral has signed early partners across several industries:
ASML — the company that makes the machines that make computer chips
European Space Agency — space operations and satellite data
Ericsson — telecommunications infrastructure
DSO & HTX Singapore — national defense and security
Reply — enterprise consulting and systems integration
Practical use cases
Government agencies could train models on policy frameworks and regulatory texts, creating AI assistants that actually understand the specific rules employees need to follow.
Software teams could train on their own codebase, so the AI writes code that matches their patterns and conventions — not generic Stack Overflow answers.
Manufacturing companies could feed engineering specs and operational manuals, creating diagnostic tools that understand their specific equipment.
The agent angle
Forge is designed to work with autonomous AI agents — not just chatbots. Mistral's own coding agent (called Vibe) can use Forge to automatically fine-tune models, generate synthetic training data, and monitor performance. The system watches key metrics to prevent "model regression" — when an updated model accidentally gets worse at something it used to do well.
Where Forge fits in the AI landscape
This puts Mistral in direct competition with OpenAI's fine-tuning API, Google's Vertex AI, and Amazon Bedrock's custom model training. The key differentiator is Mistral's positioning as a European, sovereignty-focused alternative — important for government and defense customers who need to keep data within specific jurisdictions.
Pricing hasn't been announced yet. Organizations can sign up on Mistral's website to learn more. Full details are in the official announcement.
Related Content — Get Started with Easy Claude Code | Free Learning Guides | More AI News
Stay updated on AI news
Simple explanations of the latest AI developments