Lorica Package

Introduction

This package provides functionality for interacting with Lorica Cybersecurity products. The following capabilities are currently offered:

• Attestation and OHTTP Public Key verification for Lorica AI deployment.
• OHTTP encapsulation for secure interaction with Lorica AI deployment.

Lorica AI Interaction using Requests Session

To interact with a Lorica AI deployment through a requests.Session, simply replace the object construction with lorica.Session:

import lorica
import json

# Create lorica.Session that inherits from requests.Session.
session = lorica.Session()

deployment_url = "DEPLOYMENT_URL"
lorica_api_key = "LORICA_API_KEY"

# Use session like a request.Session including response streaming support.
stream = True
resp = session.post(
    f"{deployment_url}/v1/chat/completions",
    headers={"Authorization": f"Bearer {lorica_api_key}"},
    json={
        "model": "meta-llama/Llama-3.2-3B-Instruct",
        "messages": [
            {"role": "system", "content": "You are a helpful AI assistant."},
            {"role": "user", "content": "where does the sun rise from?"},
        ],
        "temperature": 0.7,
        "max_tokens": 1024,
        "stream": stream,
    },
    stream=stream
)
resp.raise_for_status()
if stream:
    for line in resp.iter_lines(decode_unicode=True):
        if not line or not line.startswith("data: "):
            continue

        data = line[len("data: "):].strip()
        if data == "[DONE]":
            break

        chunk = json.loads(data)
        print(chunk["choices"][0]["delta"]["content"], end="", flush=True)
else:
    print(resp.json()["choices"][0]["message"]["content"])

Lorica AI Interaction using HTTPX Transport

To interact with a Lorica AI deployment through a httpx.Transport, simply replace the object construction with lorica.Transport:

import lorica
import httpx
import json

# Initialize httpx client with the lorica.Transport that inherits from httpx.Transport
httpx_client = httpx.Client(transport=lorica.Transport())

deployment_url = "DEPLOYMENT_URL"
lorica_api_key = "LORICA_API_KEY"

# Use client as normal including chunked-encoding response support.
method = "POST"
url = deployment_url + "/v1/chat/completions"
stream = True
data = {
    "model": "meta-llama/Llama-3.2-3B-Instruct",
    "messages": [
        {"role": "system", "content": "You are a helpful AI assistant."},
        {"role": "user", "content": "where does the sun rise from?"},
    ],
    "temperature": 0.7,
    "max_tokens": 1024,
    "stream": stream,
}
headers = {"Authorization": f"Bearer {lorica_api_key}"}
if stream:
    with httpx_client.stream(method, url, json=data, headers=headers) as resp:
        resp.raise_for_status()
        for line in resp.iter_lines():
            if not line or not line.startswith("data: "):
                continue

            data = line[len("data: "):].strip()
            if data == "[DONE]":
                break

            chunk = json.loads(data)
            print(chunk["choices"][0]["delta"]["content"], end="", flush=True)
else:
    resp = httpx_client.post(url, json=data, headers=headers, timeout=30)
    resp.raise_for_status()
    print(resp.json()["choices"][0]["message"]["content"])

Lorica AI Interaction using OpenAI Client

This is also applicable to clients that utilize httpx for their HTTP communication, for example openai client:

import lorica
import httpx
import openai

# Initialize httpx client with lorica.Transport that inherits from httpx.Transport
httpx_client = httpx.Client(transport=lorica.Transport())
deployment_url = "DEPLOYMENT_URL"
lorica_api_key = "LORICA_API_KEY"

# Configure OpenAI client with httpx client
client = openai.OpenAI(
    api_key=lorica_api_key,
    http_client=httpx_client,
    base_url=deployment_url + "/v1")

# Use OpenAI SDK as normal for example llama chat (including stream capability)
stream = True
completion = client.chat.completions.create(
    model="meta-llama/Llama-3.2-3B-Instruct",
    messages=[
        {"role": "system", "content": "You are a helpful AI assistant."},
        {"role": "user", "content": "where does the sun rise from?"},
    ],
    temperature=0.7,
    max_tokens=1024,
    stream=stream,
)
if stream:
    for chunk in completion:
        print(chunk.choices[0].delta.content or "", end="", flush=True)
else:
    print(completion.choices[0].message.content)

Optional Parameter Configurability

The lorica.Transport and lorica.Session class constructors take in optional parameters:

• trustee_url: In our attestation flow, a Trustee service acts as the Verifier according to the RATS attestion procedure RFC9334. The classes default to use the Trustee deployed by Lorica but this parameter can be use to override the URL to point to any other Trustee instance.
• attest: This parameter can be used to disable the attestation flow in the absence of a Trustee service. We recommend keeping attestation ON unless the network route to the deployment is trusted.

Fetching the Attested Deployment Report

The lorica.Transport and lorica.Session classes provide a get_attested_deployment_report() method that takes in a deployment's URL to output the deployment's attestation report which features attested measurements of the deployment's hardware and software. This method only outputs a report if the deployment passes all Trustee and client-side checks as part of the attestation flow. The full JSON Web Token (JWT) issued by Trustee can also be accessed via the get_attestation_token() method.