Agentic Mode

Wisent-Guard gives you unparalleled visibility into the internal thinking of the model and allows for steering, as well as easy and synthetic contrastive pair computation. What is better than the ability to do it yourself? Having an AI agent do it!

This part of the Wisent-Guard automates the process of an LLM diagnosing whether it is hallucinating or doing something bad and steering itself into a direction that allows it to create better responses.

Inspired by the reasoning architecture, the agentic mode integrates tools for the LLM to self-diagnose itself, create new contrastive pairs and generate better aligned responses. Through this, the entire Wisent-Guard system saves you time involved in optimising detailed hyper parameters, instead producing responses better aligned with your needs from the get go.