Kremis is a tool designed to validate claims made by large language models (LLMs) using empirical data. It ingests entity-attribute-value (EAV) signals and constructs a deterministic graph to classify each response into three categories: FACT (directly supported by the data), INFERENCE (derived from the data through logical reasoning), or UNKNOWN (not present in the dataset). Kremis operates without confidence scores, providing clear-cut classifications. Built with Apache 2.0 licensing, it is implemented in Rust for performance and reliability, ensuring ACID compliance via redb database technology. The tool offers an HTTP API, a command-line interface (CLI), and a MCP bridge for integration with models like Claude/Cursor.