LLMs are insecure by default. We fix that.

Trelent combines Zero Data Retention (ZDR) agreements with state of the art encryption, to give you the confidence to input sensitive information into leading AI models. That means you get better results, more control over your data, and the confidence to use these systems with your intellectual property. We encourage you to learn more about ZDR agreements under the "Security" tab of this FAQ, or by checking out our blog.

Trelent uses HYOK Encryption to encrypt and decrypt information within our services. You can learn more about HYOK under the "Security" tab of this FAQ, and by checking out our blog.

You can't. That's why Trelent is built to be used without your trust. You have full control over your data, and our job is to contractually bind the AI providers. That gets you the most secure AI chat tool in the world. Of course, we still take security incredibly seriously. We encourage you to check out our security centre, and then reach out to security@trelent.com with follow-up questions.

We currently have ZDR agreements in place with Azure OpenAI and Anthropic. We also offer Mixtral 8x7B deployed on Trelent hardware using vllm which has the same effect - no data is logged. We plan to add more providers in the future, and if you have a specific request, please feel free to send it to our sales team and we will see what we can do!

HYOK Encryption is a means of securing data whereby you, the customer, "hold" the keys at all times. That means that when we need to send that data to a user of Trelent, we first request that you decrypt it on our behalf. At any time, you can take that permission away from us or reject those requests, ensuring that you have complete control over the data, despite it being in a Trelent database.

Zero Data Retention (ZDR) agreements are legally binding contracts that Trelent signs with AI providers such as Azure OpenAI, Anthropic, and more. These contracts protect against those providers storing the inputs or outputs from your use of their models while using Trelent. It makes their models entirely stateless, so your information is only used to produce an AI response, then immediately discarded.

While the name suggests it should be truly zero, in reality it means zero request or response data. Providers may still retain metadata, such as content classification scores or information about how much a user used their models, but those are tied to anonymous identifiers. The provider is contractually bound to never save that information or otherwise expose it to employees of the provider. Under a ZDR agreement, their models become stateless.

AI providers retain data for abuse monitoring purposes. Under a ZDR agreement, they still have a right to minimal monitoring, so as to ensure their models are not being used for illegal purposes. The difference is that instead of having access to the raw input and output, they instead only have access to the results of classification models. These models "score" inputs/outputs between 0% and 100% in terms of that text's likelihood to have been a certain type of input. For example, a model being used illegally to perpetuate violence may have a high (85%-95%) violence content classification score.