Quick Start API

Prerequisites

io.net API key (get one here)
curl or Python 3.7+

Step 1: Find Models with Attestation Support

First, list available models and filter for those supporting attestation:

curl

curl -s https://api.intelligence.io.net/v1/models \
  -H "Authorization: Bearer $IO_API_KEY" | \
  jq '.data[] | select(.supports_attestation == true) | {name, model_id}'

Python

import requests

response = requests.get(
    "https://api.intelligence.io.net/v1/models",
    headers={"Authorization": f"Bearer {IO_API_KEY}"}
)

models = response.json()["data"]
attestation_models = [m for m in models if m.get("supports_attestation")]

for model in attestation_models:
    print(f"{model['name']}: {model['model_id']}")

Note the model_id (UUID) for a model you want to use.

Step 2: Get Attestation Report

Before running inference, request an attestation report to verify the GPU machine. Generate a unique nonce (random string) for freshness verification.

curl

# Generate a random nonce
NONCE=$(openssl rand -hex 16)

# Request attestation report
curl -X POST https://api.intelligence.io.net/v1/private/attestation \
  -H "Authorization: Bearer $IO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model_id": "YOUR_MODEL_UUID_HERE",
    "nonce": "'$NONCE'"
  }'

Python

import secrets
import requests

# Generate a random nonce
nonce = secrets.token_hex(16)

response = requests.post(
    "https://api.intelligence.io.net/v1/private/attestation",
    headers={
        "Authorization": f"Bearer {IO_API_KEY}",
        "Content-Type": "application/json"
    },
    json={
        "model_id": "YOUR_MODEL_UUID_HERE",
        "nonce": nonce
    }
)

attestation = response.json()
# Nonce is padded to 64 hex characters - verify prefix matches
print(f"Nonce verified: {attestation['nonce'].startswith(nonce)}")
print(f"Signing address: {attestation['signing_address']}")

Example Response

{
  "nonce": "87ebbef3ceb69d2d6d7edc1b05c42ad900000000000000000000000000000000",
  "gpu": {
    "nonce": "87ebbef3ceb69d2d6d7edc1b05c42ad900000000000000000000000000000000",
    "arch": "HOPPER",
    "evidence_list": [
      {
        "evidence": "<base64-encoded attestation evidence>",
        "certificate": "<base64-encoded certificate chain>"
      }
    ],
    "claims_version": "3.0"
  },
  "cpu": {
    "quote": "<hex-encoded CPU attestation quote>"
  },
  "signing_address": "0xf52373547CAa0EeCB0fcD34042D7518E79aA80cC",
  "image_digest": "sha256:cf47db862b96b243e077a80ee51afa2c007604bf3c648232d42144947e56c339"
}

Save the signing_address - you’ll use it to verify response signatures. Verify the image_digest - compare with the expected digest published in the latest official release to confirm the running container hasn’t been tampered with.

Step 3: Run Confidential Inference

Now run inference using the confidential completions endpoint. The request format is OpenAI-compatible.

curl

curl -X POST https://api.intelligence.io.net/v1/private/completions \
  -H "Authorization: Bearer $IO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "meta-llama/Llama-3.3-70B-Instruct",
    "messages": [
      {"role": "user", "content": "What is confidential computing?"}
    ],
    "max_tokens": 5000
  }' \
  -i  # Include response headers

Python

response = requests.post(
    "https://api.intelligence.io.net/v1/private/completions",
    headers={
        "Authorization": f"Bearer {IO_API_KEY}",
        "Content-Type": "application/json"
    },
    json={
        "model": "meta-llama/Llama-3.3-70B-Instruct",
        "messages": [
            {"role": "user", "content": "What is confidential computing?"}
        ],
        "max_tokens": 5000
    }
)

# Get the response body
completion = response.json()
print(completion["choices"][0]["message"]["content"])

# Get signature headers for verification
signature_headers = {
    "text": response.headers.get("text"),
    "signature": response.headers.get("signature"),
    "signing_address": response.headers.get("signing_address"),
    "signing_algo": response.headers.get("signing_algo"),
    "image_digest": response.headers.get("image_digest"),
}

Response Headers

The response includes signature headers for verification:

text: <signed content>
signature: <cryptographic signature>
signing_address: 0x1234...abcd
signing_algo: ecdsa
image_digest: sha256:...c00

Step 4: Verify the Response Signature

Verify that the response came from the attested machine by checking the signature.

Python (with eth_account)

from eth_account.messages import encode_defunct
from eth_account import Account

def verify_response(signature_headers, expected_signing_address):
    """Verify the response signature matches the attested machine."""

    text = signature_headers["text"]
    signature = signature_headers["signature"]
    signing_address = signature_headers["signing_address"]

    # Verify signing address matches attestation
    if signing_address.lower() != expected_signing_address.lower():
        raise ValueError("Signing address does not match attestation!")

    # Verify signature
    message = encode_defunct(text=text)
    recovered_address = Account.recover_message(message, signature=signature)

    if recovered_address.lower() != signing_address.lower():
        raise ValueError("Signature verification failed!")

    return True

# Verify using the signing_address from Step 2
signing_address_from_attestation = attestation["signing_address"]
verify_response(signature_headers, signing_address_from_attestation)
print("Response verified!")

Complete Example

Here’s a complete Python script that performs verified confidential inference:

import secrets
import requests
from eth_account.messages import encode_defunct
from eth_account import Account

IO_API_KEY = "your-api-key"
BASE_URL = "https://api.intelligence.io.net/v1"
PRIVATE_URL = "https://api.intelligence.io.net/v1/private"

def get_attestation_models():
    """Get models that support attestation."""
    response = requests.get(
        f"{BASE_URL}/models",
        headers={"Authorization": f"Bearer {IO_API_KEY}"}
    )
    models = response.json()["data"]
    return [m for m in models if m.get("supports_attestation")]

def get_attestation(model_id: str, nonce: str):
    """Get attestation report for a model."""
    response = requests.post(
        f"{PRIVATE_URL}/attestation",
        headers={
            "Authorization": f"Bearer {IO_API_KEY}",
            "Content-Type": "application/json"
        },
        json={"model_id": model_id, "nonce": nonce}
    )
    response.raise_for_status()
    return response.json()

def confidential_completion(model: str, messages: list):
    """Run confidential inference and return response with headers."""
    response = requests.post(
        f"{PRIVATE_URL}/completions",
        headers={
            "Authorization": f"Bearer {IO_API_KEY}",
            "Content-Type": "application/json"
        },
        json={"model": model, "messages": messages}
    )
    response.raise_for_status()

    return {
        "body": response.json(),
        "signature_headers": {
            "text": response.headers.get("text"),
            "signature": response.headers.get("signature"),
            "signing_address": response.headers.get("signing_address"),
            "signing_algo": response.headers.get("signing_algo"),
			"image_digest": response.headers.get("image_digest"),
        }
    }

def verify_signature(text: str, signature: str, expected_address: str):
    """Verify response signature."""
    message = encode_defunct(text=text)
    recovered = Account.recover_message(message, signature=signature)
    return recovered.lower() == expected_address.lower()

# Main flow
if __name__ == "__main__":
    # 1. Find an attestation-enabled model
    models = get_attestation_models()
    model = models[0]
    print(f"Using model: {model['name']}")

    # 2. Get attestation with fresh nonce
    nonce = secrets.token_hex(16)
    attestation = get_attestation(model["model_id"], nonce)

    # Verify nonce freshness (nonce is padded to 64 hex chars)
    assert attestation["nonce"].startswith(nonce), "Nonce mismatch!"
    signing_address = attestation["signing_address"]
    print(f"Attestation verified, signing address: {signing_address}")

    # 3. Run confidential inference
    result = confidential_completion(
        model=model["name"],
        messages=[{"role": "user", "content": "Hello, explain TEE in one sentence."}]
    )

    # 4. Verify response signature
    headers = result["signature_headers"]
    assert headers["signing_address"].lower() == signing_address.lower(), \
        "Signing address mismatch!"

    assert verify_signature(
        headers["text"],
        headers["signature"],
        signing_address
    ), "Signature verification failed!"

    print("Response verified!")
    print(result["body"]["choices"][0]["message"]["content"])

What’s Next

Confidential Chat - Use the web interface for private conversations
Verification Guide - Deep dive into verifying attestation and signatures

Getting Started with IO Intelligence

IO Intelligence Payments

Confidential Inference

Integrations

Quick Start API

Prerequisites

Step 1: Find Models with Attestation Support

curl

Python

Step 2: Get Attestation Report

curl

Python

Example Response

Step 3: Run Confidential Inference

curl

Python

Response Headers

Step 4: Verify the Response Signature

Python (with eth_account)

Complete Example

What’s Next

​Prerequisites

​Step 1: Find Models with Attestation Support

​curl

​Python

​Step 2: Get Attestation Report

​curl

​Python

​Example Response

​Step 3: Run Confidential Inference

​curl

​Python

​Response Headers

​Step 4: Verify the Response Signature

​Python (with eth_account)

​Complete Example

​What’s Next

Prerequisites

Step 1: Find Models with Attestation Support

curl

Python

Step 2: Get Attestation Report

curl

Python

Example Response

Step 3: Run Confidential Inference

curl

Python

Response Headers

Step 4: Verify the Response Signature

Python (with eth_account)

Complete Example

What’s Next