Usage Guide

Prompt Sentinel can be integrated into your LLM pipelines to protect sensitive data. Below are examples of how to use Prompt Sentinel in different scenarios.

Decorating an LLM Function Call

You can use the @sentinel decorator to sanitize messages before they reach the LLM and decode responses after processing.

@sentinel(detector=LLMSecretDetector(...))
def call_llm(messages):
    # Call the LLM with sanitized messages
    return response

Wrapping an Entire Class

Wrap an entire class to automatically sanitize and decode messages for specified methods.

InstrumentedClass = instrument_model_class(BaseChatModel, detector=LLMSecretDetector(...), methods_to_wrap=['invoke', 'ainvoke', 'stream', 'astream'])
llm = InstrumentedClass()
response = llm.invoke(messages)

Example: Simple LLM Call

Refer to simple_llm_call_example.py for a straightforward example of making an LLM call with Prompt Sentinel.

Example: Azure Client

Refer to azure_client_example.py for usage with Azure.

Example: Local VLLM Detector

Refer to local_vllm_detector_example.py for a local VLLM detector.

Example: Local HF Detector

Refer to local_hf_detector_example.py for a local Hugging Face detector.

Example: Async Azure Calls

Refer to azure_async_example.py for asynchronous Azure calls.

Example: LiteLLM

Refer to litellm_example.py for LiteLLM usage. `