swarmauri_llm_perplexity 0.11.0.dev1

Provider-specific Swarmauri import package for Perplexity chat, streaming, async, batch, and citation-aware LLM workflows.

Swarmauri Perplexity LLM

swarmauri_llm_perplexity provides the provider-specific Swarmauri import package for PerplexityModel. It wraps Perplexity's chat completion API so Swarmauri applications can use the tracked sonar and r1-1776 model families through the standard Conversation workflow.

The runtime targets https://api.perplexity.ai/chat/completions, supports sync, async, streaming, and batch generation, and exposes controls such as return_citations, top_p, top_k, presence_penalty, and frequency_penalty.

Why Use This Package?

• Keep Perplexity-specific imports explicit in Swarmauri applications.
• Use Swarmauri Conversation objects against Perplexity's hosted chat completion API.
• Stream Perplexity responses while preserving the same component model used across other provider packages.
• Access Perplexity's current repo-tracked sonar and reasoning model families from a dedicated package boundary.

FAQ

What does swarmauri_llm_perplexity install?

It installs PerplexityModel under swarmauri.llms.

Which Perplexity model families are tracked?

The current runtime allowlist includes sonar-deep-research, sonar-reasoning-pro, sonar-reasoning, sonar-pro, sonar, and r1-1776.

Does the package support streaming?

Yes. PerplexityModel supports predict, apredict, stream, astream, batch, and abatch.

Can it request citations?

Yes. The runtime exposes a return_citations option so the provider can include citation data when supported by the selected model.

Can I set both top_p and top_k?

No. The runtime rejects that combination and raises a ValueError.

Where should I verify current model and pricing details?

Use the Perplexity section in docs/LLM_PROVIDER_MODEL_PRICING_LINKS.md, which links to the current model cards and pricing documentation.

Features

• PerplexityModel import package for hosted Perplexity chat completion workflows.
• Sync, async, streaming, and batch generation support.
• Optional citation-return behavior for supported Perplexity responses.
• Usage metadata capture through Swarmauri UsageData when enabled.
• Compatibility with Python 3.10, 3.11, 3.12, 3.13, and 3.14.

Installation

uv add swarmauri_llm_perplexity

pip install swarmauri_llm_perplexity

Usage

Set PPLX_API_KEY in your environment before creating the model.

Chat Completion

import os

from swarmauri_llm_perplexity import PerplexityModel
from swarmauri_standard.conversations.Conversation import Conversation
from swarmauri_standard.messages.HumanMessage import HumanMessage

conversation = Conversation()
conversation.add_message(HumanMessage(content="Summarize Swarmauri in two sentences."))

model = PerplexityModel(
    api_key=os.environ["PPLX_API_KEY"],
    name="sonar",
)
result = model.predict(conversation=conversation, max_tokens=200)

print(result.get_last().content)

Streaming

import os

from swarmauri_llm_perplexity import PerplexityModel
from swarmauri_standard.conversations.Conversation import Conversation
from swarmauri_standard.messages.HumanMessage import HumanMessage

conversation = Conversation()
conversation.add_message(HumanMessage(content="List three differences between hosted and local inference."))

model = PerplexityModel(
    api_key=os.environ["PPLX_API_KEY"],
    name="sonar-reasoning",
)

for token in model.stream(conversation=conversation, return_citations=True):
    print(token, end="", flush=True)

Async

import asyncio
import os

from swarmauri_llm_perplexity import PerplexityModel
from swarmauri_standard.conversations.Conversation import Conversation
from swarmauri_standard.messages.HumanMessage import HumanMessage


async def main() -> None:
    conversation = Conversation()
    conversation.add_message(HumanMessage(content="Explain retrieval-augmented generation briefly."))

    model = PerplexityModel(
        api_key=os.environ["PPLX_API_KEY"],
        name="sonar-pro",
    )
    result = await model.apredict(conversation=conversation, return_citations=True)
    print(result.get_last().content)


# asyncio.run(main())

Examples

• Use sonar for general-purpose hosted Perplexity chat workflows.
• Use sonar-reasoning or sonar-reasoning-pro when you want the reasoning-oriented family explicitly.
• Use return_citations=True when the downstream experience should expose provider citations alongside the generated response.

Related Packages

• swarmauri_llm_openai
• swarmauri_llm_mistral
• swarmauri_llm_groq
• swarmauri_llm_hyperbolic
• swarmauri_llm_deepinfra
• swarmauri_llm_playht

Foundational Swarmauri Packages

• swarmauri
• swarmauri_core
• swarmauri_base
• swarmauri_standard

More Documentation

• LLM provider model and pricing links
• Perplexity model cards
• Perplexity pricing

Best Practices

• Keep PPLX_API_KEY in environment variables or a secret manager.
• Choose the Perplexity model family intentionally, especially when switching between search-style and reasoning-oriented variants.
• Do not set both top_p and top_k in the same request.
• Use citation-return mode only when the consuming application is prepared to surface or process citation metadata.

License

Apache-2.0