Perplexity LangChain Integration

Common errors

• ModuleNotFoundError: No module named 'langchain_perplexity'

  cause The `langchain-perplexity` package is not installed in your Python environment.
  fix Run `pip install langchain-perplexity` to install the package.

• ValueError: Did not find perplexity_api_key, please add an environment variable `PERPLEXITY_API_KEY` or pass `perplexity_api_key` as a named parameter.

  cause The Perplexity API key is not provided through an environment variable or directly in the `ChatPerplexity` constructor.
  fix Set the environment variable: `export PERPLEXITY_API_KEY="YOUR_API_KEY"` (or `os.environ["PERPLEXITY_API_KEY"] = "YOUR_API_KEY"` in Python code), or pass it during initialization: `ChatPerplexity(perplexity_api_key="YOUR_API_KEY", model="sonar")`.

• json.decoder.JSONDecodeError: Expecting value: line X column Y (char Z) (or similar JSON parsing error with structured output)

  cause Perplexity models, particularly reasoning-capable ones, might embed non-JSON content (like internal thought processes in `<think>` tags) within their response, which interferes with LangChain's `with_structured_output` parsing.
  fix Temporarily avoid using `with_structured_output` with Perplexity models if this issue occurs. Inspect the raw model output for extraneous text. Consider implementing a custom output parser that pre-processes the model's text to remove non-JSON elements before parsing.

Warnings

• breaking The LangChain ecosystem, including `langchain-core`, undergoes frequent updates and breaking changes. This may lead to needing to update import paths, class names, or method signatures in your `langchain-perplexity` integration code to maintain compatibility.
  Fix: Regularly consult the official LangChain migration guides and changelogs. Update your code to use the LangChain Expression Language (LCEL) and `Runnable` interface where applicable for future-proofing.
• gotcha Perplexity API key must be correctly configured. The library expects the API key to be set either as an environment variable named `PERPLEXITY_API_KEY` (or `PPLX_API_KEY` for older patterns) or passed directly to the `ChatPerplexity` constructor via the `perplexity_api_key` parameter.
  Fix: Ensure your environment variable `PERPLEXITY_API_KEY` is set: `export PERPLEXITY_API_KEY='your_api_key_here'` or `os.environ["PERPLEXITY_API_KEY"] = "your_api_key_here"`. Alternatively, instantiate `ChatPerplexity(perplexity_api_key="your_api_key_here", ...)`.
• gotcha Using `ChatPerplexity` with LangChain's `with_structured_output` functionality can fail if Perplexity models (especially 'reasoning' models) include their intermediate thought processes (e.g., text wrapped in `<think>...</think>`) in the output, breaking JSON parsing.
  Fix: If structured output is critical, inspect the raw output for `<think>` tags. You may need to implement a custom output parser that explicitly strips these tags before attempting JSON deserialization, or avoid using `with_structured_output` with models that exhibit this behavior.

Install

• pip install langchain-perplexity Install stable version

Imports

• ChatPerplexity

  from langchain_perplexity import ChatPerplexity

• PerplexitySearchRetriever

  from langchain_perplexity.retrievers import PerplexitySearchRetriever

• PerplexitySearchResults

  from langchain_perplexity.tools import PerplexitySearchResults

• WebSearchOptions

  from langchain_perplexity import WebSearchOptions

Quickstart

This quickstart demonstrates how to set up and use `ChatPerplexity` for conversational AI with real-time web search capabilities. It includes setting the API key and invoking a simple chat chain.

import os
from langchain_perplexity import ChatPerplexity
from langchain_core.prompts import ChatPromptTemplate

# Set your Perplexity API key as an environment variable
# It's recommended to load this from a .env file or similar in production
os.environ["PERPLEXITY_API_KEY"] = os.environ.get("PERPLEXITY_API_KEY", "your_perplexity_api_key_here")

# Initialize the chat model
chat = ChatPerplexity(model="sonar")

# Define a prompt template
prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a helpful assistant."),
    ("human", "{input}")
])

# Create a chain
chain = prompt | chat

# Invoke the chain with a question
response = chain.invoke({"input": "What breakthroughs in fusion energy have been announced this year?"})
print(response.content)

# Example with Pro Search (requires 'sonar-pro' model and WebSearchOptions)
# from langchain_perplexity import WebSearchOptions
# pro_chat = ChatPerplexity(
#     model="sonar-pro",
# #    web_search_options=WebSearchOptions(search_type="pro") # Uncomment if specific search_type needed
# )
# pro_response = pro_chat.invoke("How does the electoral college work?")
# print(pro_response.content)