About Granite Guardian
The Granite Guardian family is a collection of models designed to judge if the input prompts and the output responses of an LLM based system meet specified criteria. The models come pre-baked with certain criteria including but not limited to: jailbreak attempts, profanity, and hallucinations related to tool calls and retrieval augmented generation in agent-based systems.
Additionally, the models also allow users to bring their own criteria and tailor the judging behavior to specific use-cases. The README lists the license as Apache 2.0.
Drafted from github.com · not yet confirmed by IBM
How it works
Granite Guardian is IBM's family of open guardrail models, built on the Granite LLMs. They score prompts and responses for harm categories (harm, bias, jailbreaking, profanity, violence, sexual content) and also for RAG specific problems like whether an answer is grounded in the retrieved context and whether a tool call matches the user's request. That second set is what separates it from most safety classifiers.
You run it yourself or through IBM's watsonx. It's aimed at enterprises deploying LLM agents who need a check that understands agentic failures, not only toxic text.
From the project
From the project
Reviews
Be the first to review it
AnonymousKnow a team weighing Granite Guardian? Send them this page.