Enables estimating LLM hallucination and confabulation by clustering sampled responses into semantic equivalence classes and computing semantic entropy. It provides a plug-and-play validation filter for agent pipelines using the MCP protocol.
Enables comparison of responses from multiple LLMs (OpenAI, Anthropic, Gemini) to the same prompt, returning a validated divergence score based on sentence embeddings.
Real-time conversation health monitor for AI agents that tracks structural dynamics like semantic drift and information gain using deterministic arithmetic, without calling an LLM.
Enables hallucination detection for AI assistants by providing tools to assess whether responses are grounded in source material or follow grounded patterns.
Embedding space analysis toolkit that measures semantic drift, traces trajectory dynamics, and projects passages onto caller-defined semantic axes, exposed as MCP tools for agentic integration.
