Implementation:Langchain ai Langchain ChatPerplexity

Knowledge Sources	Langchain_ai_Langchain
Domains	LLM Integration, Chat Models, Search-Augmented Generation
Last Updated	2026-02-11 00:00 GMT

Overview

LangChain chat model wrapper around the Perplexity AI Chat Completions API, providing search-augmented language model capabilities.

Description

ChatPerplexity is a class in the langchain-perplexity partner package that extends BaseChatModel from langchain-core. It wraps the Perplexity AI API to provide chat completions with built-in web search capabilities, citation tracking, and support for specialized search modes (academic, SEC filings, web). The class supports synchronous and asynchronous generation, streaming, structured output via JSON schema, and Perplexity-specific features such as reasoning steps, related questions, image/video responses, and search result metadata.

Usage

Import ChatPerplexity when building LangChain applications that need Perplexity AI's search-augmented chat models with real-time web knowledge.

Code Reference

Source Location

• Repository: Langchain_ai_Langchain
• File: libs/partners/perplexity/langchain_perplexity/chat_models.py
• Lines: 1-816

Signature

class ChatPerplexity(BaseChatModel):
    client: Any = Field(default=None, exclude=True)
    async_client: Any = Field(default=None, exclude=True)
    model: str = "sonar"
    temperature: float = 0.7
    model_kwargs: dict[str, Any] = Field(default_factory=dict)
    pplx_api_key: SecretStr | None = Field(
        default_factory=secret_from_env("PPLX_API_KEY", default=None), alias="api_key"
    )
    request_timeout: float | tuple[float, float] | None = Field(None, alias="timeout")
    max_retries: int = 6
    streaming: bool = False
    max_tokens: int | None = None
    search_mode: Literal["academic", "sec", "web"] | None = None
    reasoning_effort: Literal["low", "medium", "high"] | None = None
    ...

Import

from langchain_perplexity import ChatPerplexity

I/O Contract

Inputs

Name	Type	Required	Description
model	str	No	Model name to use. Defaults to "sonar".
temperature	float	No	Sampling temperature. Defaults to 0.7.
pplx_api_key	SecretStr or None	No	Perplexity API key. Reads from PPLX_API_KEY environment variable if not set.
max_tokens	int or None	No	Maximum number of tokens to generate.
streaming	bool	No	Whether to stream results. Defaults to False.
max_retries	int	No	Maximum retries on failure. Defaults to 6.
request_timeout	float or tuple or None	No	Timeout for API requests.
search_mode	Literal["academic", "sec", "web"] or None	No	Specialized search mode for content.
reasoning_effort	Literal["low", "medium", "high"] or None	No	Reasoning effort level.
language_preference	str or None	No	Language preference for responses.
search_domain_filter	list[str] or None	No	List of domains to filter search results (max 20).
return_images	bool	No	Whether to return images. Defaults to False.
return_related_questions	bool	No	Whether to return related questions. Defaults to False.
search_recency_filter	Literal["day", "week", "month", "year"] or None	No	Filter search results by recency.
disable_search	bool	No	Whether to disable web search entirely. Defaults to False.
web_search_options	WebSearchOptions or None	No	Configuration for web search behavior.
media_response	MediaResponse or None	No	Media response configuration.

Outputs

Name	Type	Description
ChatResult	ChatResult	Contains AIMessage with content, usage_metadata, response_metadata, and additional_kwargs (citations, images, videos, reasoning_steps, related_questions, search_results).
ChatGenerationChunk	ChatGenerationChunk	When streaming, yields chunks with incremental content and metadata.

Key Methods

Method	Description
_generate(messages, stop, run_manager, **kwargs)	Synchronous generation. Delegates to streaming if streaming=True.
_agenerate(messages, stop, run_manager, **kwargs)	Async generation. Delegates to async streaming if streaming=True.
_stream(messages, stop, run_manager, **kwargs)	Synchronous streaming via Perplexity client.
_astream(messages, stop, run_manager, **kwargs)	Async streaming via Perplexity async client.
with_structured_output(schema, method, include_raw, strict, **kwargs)	Returns a Runnable that produces structured output matching the given schema. Only supports "json_schema" method.

Usage Examples

Basic Usage

from langchain_perplexity import ChatPerplexity

model = ChatPerplexity(model="sonar", temperature=0.7)
messages = [("system", "You are a chatbot."), ("user", "Hello!")]
response = model.invoke(messages)
print(response.content)

Streaming

from langchain_perplexity import ChatPerplexity

model = ChatPerplexity(model="sonar", streaming=True)
for chunk in model.stream([("user", "What is quantum computing?")]):
    print(chunk.content, end="")

Structured Output

from pydantic import BaseModel
from langchain_perplexity import ChatPerplexity

class StructuredOutput(BaseModel):
    role: str
    content: str

model = ChatPerplexity(model="sonar")
structured = model.with_structured_output(StructuredOutput)
result = structured.invoke([("user", "Describe your role.")])

Response Metadata

Perplexity responses include rich metadata in additional_kwargs on the AIMessage:

• citations: List of cited sources from web search.
• images: Image results when return_images is enabled.
• videos: Video results when media_response includes videos.
• related_questions: Related questions when return_related_questions is enabled.
• search_results: Raw search results.
• reasoning_steps: Reasoning step details for reasoning-capable models.

The response_metadata includes:

• model_name: The model used for the response.
• num_search_queries: Number of search queries executed.
• search_context_size: Size of the search context used.