Frequently Asked Questions

What models do you support?

In our managed cloud (shared) platform, we support the best open LLMs currently available including those from the Llama 3.1, Mistral, and deepseek families. In single-tenant deployments, you can tailor the model mix to fit the needs of your company!

How does Prediction Guard protect our data?

We don't store, log, or cache any prompt data flowing through our system. If you are self-hosting Prediction Guard, you can even keep the deployment air-gapped. We couldn't train models on your data if we wanted to! We also enable you to filter PII, anonymize data injected into prompts, filter prompt injections, and more in accordance with NIST and OWASP recommendations. For customers leveraging self-hosting, data doesn't even leave your network(s).

Can I use LangChain and LlamaIndex?

Yes! We love LangChain and LlamaIndex! In fact, we have officially supported PG integrations in the LangChain codebase.

How can I integrate my company data?

95% of use cases can be solved using a methodology called "retrieval augmentation". This is great, because all data can stay within your company's infra and be injected into the privately hosted models without that data flowing back to the creators of those models.

Are you HIPAA compliant?

Yes. We can sign a BAA.

Back to top