Blindfold AI

As AI shifts to agents that touch credentials, patient records, and source code, that gap stopped being academic. We wanted to build the thing that lets those teams finally say yes: inference where the prompt, the response, and the model weights are never visible to the OS, the cloud provider, or even us — with a receipt to prove it.

Blindfold AI is a confidential inference gateway. You point it at your app with the same OpenAI-compatible API you already use, and every request runs inside a hardware-encrypted Trusted Execution Environment. Prompts, responses, and model weights stay encrypted even during processing. The twist that makes it trustworthy instead of just "trust us": every response ships with a cryptographic attestation — a hardware-backed receipt proving exactly which model, container, and runtime handled the request. Your compliance team can verify that privacy held without ever reading the data itself.

Built With

• amazon-web-services
• openai
• python
• react
• rust
• vercel
• vllm