​

Sentinel: Complete Data Privacy

Sentinel is an open-source, low-latency tokenization proxy designed to achieve complete data privacy and localization when interacting with external servers. It acts as a secure intermediary, ensuring that sensitive data never leaves your environment, while still allowing seamless interaction with Inferable’s services.

With Sentinel, you gain:

• Complete masking of all data sent to Inferable’s control-plane
• Guarantee that private data never leaves your environment
• Zero visibility of sensitive information by Inferable
• Ability to audit all traffic between your services and the Inferable API

​

How It Works

Sentinel acts as a sophisticated privacy barrier for your data:

1. Sensitive data is replaced with tokens before leaving your environment.
2. Only tokenized data is forwarded to Inferable’s servers.
3. Responses containing tokens are intercepted and original data is restored before reaching your services.

This process ensures that Inferable never sees or stores your sensitive information.

​

Deployment

To maintain the highest level of data privacy:

1. Sentinel must be deployed within your own VPC, alongside your services.
2. You need to configure your own LLMs within your secure environment.
3. The inference layer (LLMs) must also be deployed within your VPC.

​

Without Sentinel

Orchestration: Your services talk directly to the Inferable API.

Inference: All model calls are made directly inside the Control Plane.

​

With Sentinel

Orchestration: Your services communicate exclusively with Sentinel, which tokenizes all outgoing data and detokenizes incoming data.

Inference: The Inferable Orchestrator forwards model calls to Sentinel, which then masks/unmasks data as necessary before forwarding for inference, keeping all sensitive data within your environment.

By implementing Sentinel, you ensure that Inferable never has access to or visibility of your sensitive data, and a verifieable audit trail of all traffic between your services and the Inferable API.