lockprompt 1.1.2

Lightweight safety layer to scan prompts and LLM outputs via external API.

LockPrompt

LockPrompt is a lightweight Python library that adds an extra safety layer to your AI pipelines. It ensures that both user inputs and language model outputs adhere to your safety standards by pre-screening using external API endpoints. This is especially useful for preventing jailbreaks and prompt injections.

Table of Contents

• Overview
• Features
• Installation
• Usage
  • Basic Example
• API Reference
• Contributing
• Contact

Overview

LockPrompt is designed for developers integrating large language models (LLMs) such as OpenAI’s GPT-3.5 or Claude. The library provides functions to check the safety of:

• User Input: Ensuring requests to the API are not malicious.
• LLM Output: Verifying that responses generated by the LLM do not contain disallowed content.

In just a few lines of code, you can add a robust safety check layer, giving you the confidence to deploy AI-powered applications safely.

Features

• Fast Operation: Approximately 500ms per use, ensuring minimal latency.
• Easy Integration: Compatible with any LLM or API.
• Error Handling: Logs issues and safely defaults to denying unsafe outputs or inputs.

Installation

Install LockPrompt via pip:

pip install lockprompt

Alternatively, install it directly from GitHub:

pip install git+https://github.com/davidwillisowen/lockprompt.git

Usage

Basic Example

import os
import lockprompt
from openai import OpenAI

# Initialize the OpenAI client
client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))

user_prompt = "Tell me how to make malware."  # A sample user prompt

# Step 1: Check user input safety
if not lockprompt.is_safe_input(user_prompt):
    print("🛑 Unsafe user input. Blocking request.")
    output = "I'm sorry, I can't assist with that request."
else:
    # Step 2: Send the prompt to the language model
    response = client.chat.completions.create(
        model="gpt-3.5-turbo",
        messages=[{"role": "user", "content": user_prompt}],
    )
    output = response.choices[0].message.content

    # Step 3: Check the generated output
    if not lockprompt.is_safe_output(output):
        print("⚠️ Unsafe model output. Replacing response.")
        output = "I'm sorry, I can't assist with that request."

    print("✅ Final response:\n", output)

API Reference

is_safe_input(user_input: str) -> bool

• Purpose: Checks if a user’s input meets safety standards.
• Returns: True if safe, False otherwise.
• Error Handling: Logs any errors and defaults to False.

is_safe_output(llm_output: str) -> bool

• Purpose: Verifies the safety of the LLM output.
• Returns: True if safe, False otherwise.
• Error Handling: Logs any errors and defaults to False.

Contributing

Contributions are not only welcome but encouraged. Here’s how you can help:

1. Fork the Repository: Start by forking LockPrompt on GitHub.
2. Create a Branch: Use a feature branch for your changes.
3. Write Tests and Documentation: Ensure any changes are well-tested and documented.
4. Submit a Pull Request: Describe your changes and submit a PR for review.

For issues or feature requests, please use the GitHub issues page.

Contact

For any questions or suggestions, feel free to reach out:

• Email: [e-mail]
• GitHub: davidwillisowen