Bespoke-Minicheck, a grounded factuality checking model built by Bespoke Labs, is now available in Ollama. It fact-checks responses generated by other LLMs by comparing a claim against a source document and outputting Yes or No depending on whether the document supports the claim. It's particularly useful as a post-processing step in RAG applications to catch hallucinations. Getting started requires running 'ollama run bespoke-minicheck' and providing a Document and Claim in the prompt. Example implementations for RAG-based and simple fact-checking use cases are available on GitHub.
Questions this post answers
How do I detect hallucinations in RAG application responses using Ollama?
Use the Bespoke-Minicheck model, now available in Ollama, which fact-checks a generated claim against the source document retrieved for a RAG query. Run it with 'ollama run bespoke-minicheck', then prompt it with a Document and Claim pair; it outputs Yes if the document supports the claim or No if it doesn't, letting you catch hallucinations as a post-processing step. daily.dev surfaces practical guides like this for developers hardening RAG pipelines against hallucinations.
What model can I use to fact-check whether an LLM claim is grounded in a source document?
Bespoke-Minicheck, developed by Bespoke Labs and distributed through Ollama, is built specifically for grounded factuality checking. It takes a Document and a Claim as input and outputs a simple Yes or No verdict on whether the claim is supported by the document, making it suitable for automated hallucination detection pipelines. Developers evaluating factuality-checking tools can track new model releases like this via daily.dev.