FactGenie is an open-source, self-hosted web application for annotating errors in large-language-model outputs. The project, maintained by the Institute of Formal and Applied Linguistics at Charles University, supports both human annotation and API-based evaluation.
According to its GitHub documentation, the tool can collect annotations from crowdworkers, visualize labeled spans and calculate statistics for analysis. It is designed for organizations that already have datasets and model outputs; it does not itself recruit annotators or supply an AI model.
The project’s January 2026 release notes list version 1.2.1. Users can install the package with Python, run a local server and consult the project’s wiki for setup, data-management and analysis guidance.
Source: FactGenie GitHub repository, version information current Jan. 15, 2026.
