llm-perplexity 2026.2.1

LLM access to pplx-api 3 by Perplexity Labs

llm-perplexity

LLM access to pplx-api 3 by Perplexity Labs

Installation

Install this plugin in the same environment as LLM.

llm install llm-perplexity

Usage

First, set an API key for Perplexity AI:

llm keys set perplexity
# Paste key here

Run llm models to list the models, and llm models --options to include a list of their options.

Available Models

All Perplexity models have access to real-time web information. Here are the currently available models (as of 2026-02) from https://docs.perplexity.ai/models/model-cards:

• sonar-pro - Flagship model (200k context) - with web search
• sonar - Base model (128k context) - with web search
• sonar-deep-research - Deep research model (128k context) - with web search
• sonar-reasoning-pro - Advanced reasoning model (128k context) - with web search

Run prompts like this:

# Flagship model
llm -m sonar-pro 'Latest AI research'

# Base model
llm -m sonar 'Fun facts about walruses'

# Research and reasoning models
llm -m sonar-deep-research 'Complex research question'
llm -m sonar-reasoning-pro 'Problem solving task'

Advanced Options

The plugin supports various parameters to customize model behavior:

# Control randomness (0.0 to 2.0, higher = more random)
llm -m sonar-pro --option temperature 0.7 'Generate creative ideas'

# Nucleus sampling threshold (alternative to temperature)
llm -m sonar-pro --option top_p 0.9 'Generate varied responses'

# Token filtering (between 0 and 2048)
llm -m sonar-pro --option top_k 40 'Generate focused content'

# Limit response length
llm -m sonar-pro --option max_tokens 500 'Summarize this article'

# Return related questions
llm -m sonar-pro --option return_related_questions true 'How does quantum computing work?'

# Use Pro Search or auto classification (requires streaming)
llm -m sonar-pro --option search_type pro 'Analyze the latest developments in quantum computing'
llm -m sonar-pro --option search_type auto 'Compare the energy efficiency of popular EVs'

# Suppress citations section and discourage inline [n] markers
llm -m sonar-pro --option include_citations false 'Latest AI research'

# Search mode: web (default), academic, or sec (SEC filings)
llm -m sonar-pro --option search_mode academic 'Recent papers on transformer architectures'
llm -m sonar-pro --option search_mode sec 'Apple quarterly earnings'

# Disable web search entirely
llm -m sonar-pro --option disable_search true 'What is the capital of France?'

# Filter search results by domain
llm -m sonar-pro --option search_domain_filter 'arxiv.org,nature.com' 'Recent AI papers'

# Filter search results by recency (hour, day, week, month, year)
llm -m sonar-pro --option search_recency_filter year 'Major events'

# Filter search results by language
llm -m sonar-pro --option search_language_filter en 'Latest tech news'

# Penalize repetition (presence_penalty: -2.0 to 2.0)
llm -m sonar-pro --option presence_penalty 1.5 'Tell me about different dog breeds'

# Penalize frequent tokens (frequency_penalty: > 0, 1.0 = no penalty)
llm -m sonar-pro --option frequency_penalty 1.5 'Write a varied paragraph about nature'

# Control reasoning effort (minimal, low, medium, high)
llm -m sonar-reasoning-pro --option reasoning_effort high 'Solve this complex math problem'

# Include images in results
llm -m sonar-pro --option return_images true 'Show me the Eiffel Tower'

# Set output language preference
llm -m sonar-pro --option language_preference es 'Tell me about AI'

# Stop sequences
llm -m sonar-pro --option stop 'END' 'Generate a list'

Using Images with Perplexity

The plugin supports sending images to Perplexity models for analysis (multi-modal input):

# Analyze an image with Perplexity
llm -m sonar-pro --option image_path /path/to/your/image.jpg 'What can you tell me about this image?'

# Ask specific questions about an image
llm -m sonar-pro --option image_path /path/to/screenshot.png 'What text appears in this screenshot?'

# Multi-modal conversation with an image
llm -m sonar-pro --option image_path /path/to/diagram.png 'Explain the process shown in this diagram'

Note: Only certain Perplexity models support image inputs. Currently the following formats are supported: PNG, JPEG, and GIF.

OpenRouter Access

You can also access these models through OpenRouter. First install the OpenRouter plugin:

llm install llm-openrouter

Then set your OpenRouter API key:

llm keys set openrouter

Use the --option use_openrouter true flag to route requests through OpenRouter:

llm -m sonar-pro --option use_openrouter true 'Fun facts about pelicans'

Development

To set up this plugin locally, first checkout the code. Then create a new virtual environment:

cd llm-perplexity
python3 -m venv venv
source venv/bin/activate

Now install the dependencies and test dependencies:

llm install -e '.[test]'

Running Tests

The test suite is comprehensive and tests all example commands from the documentation with actual API calls.

Before running tests, you need to set up your environment variables:

1. Copy the .env.example file to .env:

   cp .env.example .env
   

2. Edit the .env file and add your Perplexity API key:

   LLM_PERPLEXITY_KEY=your_perplexity_api_key_here
   

3. (Optional) If you want to test OpenRouter integration, also add your OpenRouter API key:

   LLM_OPENROUTER_KEY=your_openrouter_api_key_here
   

4. Install the package and test dependencies using one of these methods:

   Using the setup script:

   ./setup.sh
   

   Using make:

   make setup
   

   Manually:

   pip install -e .
   pip install pytest python-dotenv pillow
   

Run the tests with pytest:

# Run all tests
pytest test_llm_perplexity.py

# Using make
make test

# Run a specific test
pytest test_llm_perplexity.py::test_standard_models

Note: Running the full test suite will make real API calls to Perplexity, which may incur costs depending on your account plan.

This plugin was made after the llm-claude-3 plugin by Simon Willison.