boost.ai launches Presis, a self-hosted model for routing and safety checks in regulated AI agents

Norwegian conversational AI company boost.ai launches Presis, a self-hosted language model for classification tasks in its conversational AI platform, the company said in a press release. The fine-tuned model determines which part of an AI agent should respond, whether questions are within scope, whether answers are safe to show users and how completed conversations are reviewed.

According to boost.ai's internal testing, Presis is at least 2.5 times faster than the models it replaces, with the difference most noticeable at peak demand and in voice deployments. The company says it also reduced average token costs by 33 percent and improved accuracy across five classification features by 0.5 to 7.9 percentage points.

Presis runs on infrastructure controlled by boost.ai, reducing its reliance on external model providers for routing and safeguard checks. The model rolls out across the platform in the coming weeks, and boost.ai is evaluating a similar approach for generative features. The company serves banks and insurers including Nordea and DNB.