TurboPrivate docsCreate API account

Quickstart

Make your first model request with the Python SDK, receive an answer and close the Room. This is a complete application example; no CLI is required.

Before you start

This example uses paid inference. For another integration path, choose Direct API, JavaScript SDK or CLI for an existing tool.

1. Install

python -m pip install https://api.turboprivate.ai/sdk/python.tgz

2. Configure your credentials

Set TURBOPRIVATE_API_KEY to your key and TURBOPRIVATE_MODEL to the live model ID you selected. The example reads both from the environment. Use your shell's environment settings or a secret manager; do not put the key in the source file.

For example, in Bash (the key is entered without echo):

read -rsp 'API key: ' TURBOPRIVATE_API_KEY; echo
export TURBOPRIVATE_API_KEY
read -rp 'Live model ID: ' TURBOPRIVATE_MODEL
export TURBOPRIVATE_MODEL

The tp_live_… API key identifies your account. It is used by all three access methods. The SDK will handle Room credentials and encryption separately.

3. Send a request

Save this as example.py:

import os
from turboprivate import TurboPrivate

with TurboPrivate(
    api_key=os.environ["TURBOPRIVATE_API_KEY"],
    model=os.environ["TURBOPRIVATE_MODEL"],
) as client:
    result = client.chat.completions.create(
        messages=[{"role": "user", "content": "Say hello in one sentence."}],
    )
    print(result["choices"][0]["message"]["content"])
    print("Receipt verified:", client.last_receipt_verified)
python example.py

The SDK authenticates, allocates and verifies a Room, encrypts the request and checks the signed response. The with block attempts to close the Room when it exits, including when the request raises an error. A connection failure can leave cleanup unconfirmed; idle expiry is the fallback.

4. Check the result

You should see the model's answer and Receipt verified: True. The exact answer varies. Receipt verification checks the signed response; it does not establish host confidentiality. For an exportable record, continue to Privacy proofs.

If no model is live, stop and retry when capacity is available. For authentication, balance or connection errors, use Troubleshooting. Do not bypass trust checks or send plaintext to the public API host.

Next steps