factool

Why we included this project

Anyone who has watched ChatGPT answer confidently with something completely wrong knows the problem FacTool tackles. It is a tool-augmented framework that runs a verification pass over text a model generated, covering four kinds of output: knowledge-based QA, code, math reasoning, and scientific literature. That means one library can flag hallucinated facts, code that will not run as written, calculation slips, or citations to papers that do not exist. You install it with pip, call it from Python, and get back a structured report with individual claims, supporting evidence, and a per-claim factuality score, so it fits into an evaluation or QA pipeline rather than staying a research demo. A ChatGPT plugin gives you the same checks interactively, and the companion Halu-J model handles critique-based hallucination scoring. If you build guardrails around chat or generated content, this is a tool worth keeping close.