Run LLMs in secure GPU TEE with 99% native performance. Our OpenAI-compatible API gives you enterprise-grade privacy without sacrificing speed or ease of use.
Your data stays encrypted during processing, and you can verify the integrity of every response. This gives you enterprise-grade privacy with the simplicity of a standard API.

​
Overview

We will introduce the usage of AI API. Here is the workflow of how to use Confidential AI API, including attestation verifying and signature verifying.

​
Prerequisites Before API Requests

​
1. Add Funds

You need at least $5 in your account. Go to Dashboard and click Deposit.

​
2. Enable the API

GPU TEE API
Navigate to Dashboard → Confidential AI API and click Enable. Then create your first API key and click the key to copy.
GPU TEE API Generate Key

​
3. Choose Your Model

We support 6+ models running in GPU TEE. Click the GPU TEE checkbox to see all options:
Models in TEE

​
Make Your First Request

Use our OpenAI-compatible endpoint to chat with secure AI: 
https://api.redpill.ai/v1/chat/completions
Here’s a simple curl example:
CLI
curl -X 'POST' \
  'https://api.redpill.ai/v1/chat/completions' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer <YOUR_API_KEY>' \
  -d '{
  "messages": [
    {
      "content": "You are a helpful assistant.",
      "role": "system"
    },
    {
      "content": "What is your model name?",
      "role": "user"
    }
  ],
  "stream": true,
  "model": "phala/deepseek-chat-v3-0324"
}'
You’ll get a streaming response like this: 
data: {"id":"chatcmpl-0cdf7629fcfa4135bbdb9936e737e95c","object":"chat.completion.chunk","created":1740415146,"model":"phala/deepseek-chat-v3-0324","choices":[{"index":0,"delta":{"content":""},"logprobs":null,"finish_reason":"stop","stop_reason":128001}]}
...
data: [DONE]

​
Verify Your AI is Running Securely

Every response comes with cryptographic proof that it ran in a secure TEE. Here’s how to verify this.

​
Get the Attestation Report

Request an attestation report for any model:
CLI
curl \
  'https://api.redpill.ai/v1/attestation/report?model=phala/deepseek-chat-v3-0324' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer <YOUR_API_KEY>'
The response includes: 
{
  "signing_address": "...",
  "nvidia_payload": "...",
  "intel_quote": "...",
  "all_attestations": [
    {
      "signing_address": "...",
      "nvidia_payload": "...",
      "intel_quote": "..."
    }
  ]
}
The signing_address is a cryptographic key generated inside the TEE. It’s used to sign chat responses, proving they came from the verified hardware. The all_attestations list contains attestations from all GPU nodes in our network. Use any signing_address from this list to verify the integrity of that specific node.

​
Verify GPU Attestation

Copy the nvidia_payload and verify with NVIDIA:
CLI
curl -X POST https://nras.attestation.nvidia.com/v3/attest/gpu \
 -H "accept: application/json" \
 -H "content-type: application/json" \
 -d "<NVIDIA_PAYLOAD_FROM_ABOVE>"

​
Verify Intel TDX Attestation

Verify the Intel TDX attestation quote using the intel_quote value at the TEE Attestation Explorer.

​
Verify Chat Signatures

Every chat response includes a cryptographic signature. Here’s how to verify them.

​
Get the Signature

When you chat with an LLM, the response contains an id field with prefix start with chatcmpl-.
Use this ID to retrieve the signature within 20 minutes.
Sample Response
...
data: {"id":"chatcmpl-0cdf7629fcfa4135bbdb9936e737e95c","object":"chat.completion.chunk","created":1740415146,"model":"phala/deepseek-chat-v3-0324","choices":[{"index":0,"delta":{"content":""},"logprobs":null,"finish_reason":"stop","stop_reason":128001}]}
data: [DONE]
Use the response id to get the signature: 
# URL format
GET https://api.redpill.ai/v1/signature/{request_id}?model={model_id}&signing_algo=ecdsa
For example, with the previous response ID chatcmpl-0cdf7629fcfa4135bbdb9936e737e95c:
Get Signature Response
# Example
curl -X GET https://api.redpill.ai/v1/signature/chatcmpl-0cdf7629fcfa4135bbdb9936e737e95c?model=phala/deepseek-chat-v3-0324&signing_algo=ecdsa

# Response output
{
  "text": "65b0adb47d0450971803dfb18d0ce4af4a64d27420a43d5aad4066ebf10b81b5:e508d818744d175a62aae1a9fb3f373c075460cbe50bf962a88ac008c843dff1",
  "signature": "0xf28f537325c337fd96ae6e156783c904ca708dcd38fb8a476d1280dfc72dc88e4fcb5c3941bdd4f8fe5238a2253b975c6b02ea6a0a450b5b0f9296ab54cf24181b",
  "signing_address": "0xc51268C9b46140619CBC066A34441a6ca51F85f9",
  "signing_algo": "ecdsa"
}
Response Fields:
  • text: The message to verify
  • signature: The cryptographic signature
  • signing_address: The public key in TEE
  • signing_algo: The signature algorithm used
How do we generate text?text content is generated by SHA256(request body) + ":" + SHA256(response body)Example
NoticeSometime the api gateway may change the request body fields (i.e., model, or stream_options), so the request body sha256 may not be the same as your original request body sha256.
Signatures are stored in memory for only 20 minutes after the response is generated.

​
Verify on Etherscan

Go to Etherscan’s Signature Verification and click “Verify Signature”:
  • Address: Use the signing_address from the attestation API
  • Message: Use the text field from the signature response
  • Signature Hash: Use the signature field from the signature response
GPU TEE API Verify Signature