Confidential Service Verifier Python SDK

A pure Python SDK for fetching and verifying TEE (Trusted Execution Environment) hardware attestation reports from various providers like Tinfoil, Redpill, and Near AI. Supports both Intel TDX and Nvidia CC (Confidential Computing) GPU attestation.

1. Dependencies and Pre-requisites

To ensure reliable verification, the following dependencies are required:

• Python Runtime (uv): We use uv for modern, fast Python dependency management.
• Docker: Required for running the dstack-verifier service.
• DStack Verifier: The verification process for Redpill and Near AI apps relies on dstack-verifier, which uses QEMU internally. QEMU is essential to stably reproduce ACPI table contents and other low-level boot measurements required for TDX quote verification.

Setup

# 1. Start the dstack-verifier service
docker compose up -d

# 2. Sync dependencies
uv sync

2. Configuration

For Tinfoil verification, the SDK requires an enclave configuration file.

Important

Always ensure you are using the latest tinfoil_config.yml. The SDK looks for this file in config/tinfoil_config.yml.

To update the config:

# Fetch latest config
uv run scripts/update_config.py

Intel Trust Authority (Optional)

The SDK supports optional remote appraisal of Intel TDX quotes via Intel Trust Authority (ITA). This works with all supported providers (Tinfoil, Redpill, NearAI).

To enable this, obtain an API key from the Intel Trust Authority Portal and set it in your .env file:

INTEL_TRUST_AUTHORITY_API_KEY="your-api-key-here"

When enabled, the verification result will include an additional intel_trust_authority claim containing the appraisal details.

3. Quick Test via Server

The SDK includes a FastAPI server for testing and easy integration. This is the fastest way to verify models.

Run the Server

PYTHONPATH=. uv run python server/main.py

Quick Verification Check

You can use curl to verify any supported model.

Example: Verifying a Tinfoil Model

curl "http://localhost:8000/verify-model?provider=tinfoil&model_id=kimi-k2-thinking"

Example Output:

{
  "model_verified": true,
  "provider": "tinfoil",
  "timestamp": 1768923695.4072542,
  "hardware_type": ["INTEL_TDX"],
  "model_id": "kimi-k2-thinking",
  "claims": {
    "status": "UpToDate",
    "hw_profile": "large_1d_new"
  },
  "error": null
}

4. SDK Usage and Sample Outputs

The SDK provides a clean API for programmatic verification.

Sample Code (Python)

import asyncio
from confidential_verifier import TeeVerifier

async def main():
    verifier = TeeVerifier()

    # Verify a model directly (fetches + verifies)
    # Supports "redpill", "nearai", "tinfoil"
    result = await verifier.verify_model("redpill", "meta-llama/llama-3.3-70b-instruct")

    print(f"Model Verified: {result.model_verified}")
    print(f"Hardware: {result.hardware_type}")

    if result.model_verified:
        print(f"Claims: {result.claims}")
    else:
        print(f"Error: {result.error}")

if __name__ == "__main__":
    asyncio.run(main())

Sample Outputs

NearAI

{
  "model_verified": true,
  "provider": "nearai",
  "hardware_type": ["INTEL_TDX", "NVIDIA_CC"],
  "model_id": "openai/gpt-oss-120b",
  "request_nonce": "7299aba5...",
  "signing_address": "0x5c49f3...",
  "claims": {
    "components": {
      "gateway": { "is_valid": true, "tcb_status": "UpToDate" },
      "model": { "is_valid": true, "tcb_status": "UpToDate" }
    },
    "nvidia": { "x-nvidia-overall-att-result": true }
  }
}

Redpill

{
  "model_verified": true,
  "provider": "redpill",
  "hardware_type": ["INTEL_TDX", "NVIDIA_CC"],
  "model_id": "phala/gpt-oss-20b",
  "claims": {
    "phala": { "model_provider": "phala" },
    "nvidia": { "x-nvidia-overall-att-result": true }
  }
}

Tinfoil

{
  "model_verified": true,
  "provider": "tinfoil",
  "timestamp": 1768923695.4072542,
  "hardware_type": ["INTEL_TDX"],
  "model_id": "kimi-k2-thinking",
  "request_nonce": null,
  "signing_address": null,
  "claims": {
    "status": "UpToDate",
    "advisory_ids": [],
    "repo": "tinfoilsh/confidential-kimi-k2-thinking",
    "hw_profile": "large_1d_new"
  },
  "error": null
}

Documentation

• Tinfoil Verification Details
• Redpill Verification Details
• NearAI Verification Details
• Intel Trust Authority Details

Features

• Multi-Provider Support: Tinfoil, Redpill, and Near AI.
• Hardware Verification: Intel TDX (Local + Remote ITA) and Nvidia CC (GPU).
• Phala Cloud Integration: Native support for dstack apps on Phala.
• Resale Verification: Correctly verifies models resold between providers.
• Automated Manifests: Sigstore integration for Tinfoil.