Skip to main content

PerplexityIntegration

What it is​

A small integration client for Perplexity’s API that:

  • Calls POST /chat/completions to perform web-backed Q&A via Perplexity models.
  • Optionally exposes the search capability as LangChain StructuredTools.

Public API​

PerplexityIntegrationConfiguration​

Dataclass configuration (extends IntegrationConfiguration):

  • api_key: str — Perplexity API key (Bearer token).
  • base_url: str = "https://api.perplexity.ai" — API base URL.
  • system_prompt: str = "Be precise and concise and answer the question with sources." — default system prompt.

PerplexityIntegration​

Integration client (extends Integration):

  • __init__(configuration: PerplexityIntegrationConfiguration)
    • Initializes headers with Authorization: Bearer <api_key> and JSON content type.
  • search_web(...) -> str
    • Sends a chat completion request with web search options and returns response["choices"][0]["message"]["content"].
    • Key parameters:
      • question: str — user question.
      • system_prompt: str | None — overrides config default when provided.
      • model: str = "sonar-pro" — model name passed to Perplexity.
      • Search/web options: search_mode, search_context_size, user_location, search_recency_filter, search_domain_filter, etc.
      • Generation controls: max_tokens, temperature, top_p, top_k, presence_penalty, frequency_penalty, reasoning_effort.
    • Internally removes payload keys where the value is None, [], or {} before sending.

as_tools(configuration: PerplexityIntegrationConfiguration) -> list​

Factory returning LangChain StructuredTool objects backed by PerplexityIntegration.search_web:

  • perplexity_quick_search — uses model="sonar".
  • perplexity_search — uses model="sonar-pro".
  • perplexity_advanced_search — uses model="sonar-pro" and forces search_context_size="high".

Each tool uses a Pydantic schema for arguments:

  • question: str (required)
  • user_location: str (default "FR")
  • search_context_size: str (default "medium", validated as low|medium|high) for quick/search tools.

Configuration/Dependencies​

  • Requires:
    • requests
    • pydantic
    • naas_abi_core.integration (Integration, IntegrationConfiguration, IntegrationConnectionError)
  • Optional (only if using as_tools):
    • langchain_core.tools.StructuredTool

Usage​

Basic client usage​

from naas_abi_marketplace.ai.perplexity.integrations.PerplexityIntegration import (
    PerplexityIntegration,
    PerplexityIntegrationConfiguration,
)

config = PerplexityIntegrationConfiguration(api_key="YOUR_PERPLEXITY_API_KEY")
client = PerplexityIntegration(config)

answer = client.search_web(
    question="What is the current GDP of France? Provide sources.",
    user_location="FR",
    search_context_size="medium",
)
print(answer)

LangChain tools​

from naas_abi_marketplace.ai.perplexity.integrations.PerplexityIntegration import (
    PerplexityIntegrationConfiguration,
    as_tools,
)

tools = as_tools(PerplexityIntegrationConfiguration(api_key="YOUR_PERPLEXITY_API_KEY"))
# tools is a list of StructuredTool instances (perplexity_quick_search, perplexity_search, perplexity_advanced_search)

Caveats​

  • Errors from HTTP requests are wrapped and raised as IntegrationConnectionError.
  • search_web assumes the API response contains choices[0].message.content; missing/changed response structure will raise a KeyError.
  • search_domain_filter default is a mutable empty list ([]) in the function signature.