Perplexity Search

Perplexity Search

Overview

Perform AI-powered web searches using Perplexity models through LiteLLM and OpenRouter. Perplexity provides real-time, web-grounded answers with source citations, making it ideal for finding current information, recent scientific literature, and facts beyond the model's training data cutoff.

This skill provides access to all Perplexity models through OpenRouter, requiring only a single API key (no separate Perplexity account needed).

When to Use This Skill

Use this skill when:

• Searching for current information or recent developments (2024 and beyond)
• Finding latest scientific publications and research
• Getting real-time answers grounded in web sources
• Verifying facts with source citations
• Conducting literature searches across multiple domains
• Accessing information beyond the model's knowledge cutoff
• Performing domain-specific research (biomedical, technical, clinical)
• Comparing current approaches or technologies

Do not use for:

• Simple calculations or logic problems (use directly)
• Tasks requiring code execution (use standard tools)
• Questions well within the model's training data (unless verification needed)

Quick Start

Setup (One-time)

1. Get OpenRouter API key:

• Visit https://openrouter.ai/keys
• Create account and generate API key
• Add credits to account (minimum $5 recommended)

1. Configure environment: ```bash # Set API key export OPENROUTER_API_KEY='sk-or-v1-your-key-here'

# Or use setup script python scripts/setup_env.py --api-key sk-or-v1-your-key-here

3. **Install dependencies**:

uv pip install litellm

4. **Verify setup**:

python scripts/perplexity_search.py --check-setup

See `references/openrouter_setup.md` for detailed setup instructions, troubleshooting, and security best practices.

### Basic Usage

**Simple search:**

python scripts/perplexity_search.py "What are the latest developments in CRISPR gene editing?"

**Save results:**

python scripts/perplexity_search.py "Recent CAR-T therapy clinical trials" --output results.json

**Use specific model:**

python scripts/perplexity_search.py "Compare mRNA and viral vector vaccines" --model sonar-pro-search

**Verbose output:**

python scripts/perplexity_search.py "Quantum computing for drug discovery" --verbose

## Available Models

Access models via `--model` parameter:

- **sonar-pro** (default): General-purpose search, best balance of cost and quality
- **sonar-pro-search**: Most advanced agentic search with multi-step reasoning
- **sonar**: Basic model, most cost-effective for simple queries
- **sonar-reasoning-pro**: Advanced reasoning with step-by-step analysis
- **sonar-reasoning**: Basic reasoning capabilities

**Model selection guide:**
- Default queries → `sonar-pro`
- Complex multi-step analysis → `sonar-pro-search`
- Explicit reasoning needed → `sonar-reasoning-pro`
- Simple fact lookups → `sonar`
- Cost-sensitive bulk queries → `sonar`

See `references/model_comparison.md` for detailed comparison, use cases, pricing, and performance characteristics.

## Crafting Effective Queries

### Be Specific and Detailed

**Good examples:**
- "What are the latest clinical trial results for CAR-T cell therapy in treating B-cell lymphoma published in 2024?"
- "Compare the efficacy and safety profiles of mRNA vaccines versus viral vector vaccines for COVID-19"
- "Explain AlphaFold3 improvements over AlphaFold2 with specific accuracy metrics from 2023-2024 research"

**Bad examples:**
- "Tell me about cancer treatment" (too broad)
- "CRISPR" (too vague)
- "vaccines" (lacks specificity)

### Include Time Constraints

Perplexity searches real-time web data:
- "What papers were published in Nature Medicine in 2024 about long COVID?"
- "What are the latest developments (past 6 months) in large language model efficiency?"
- "What was announced at NeurIPS 2023 regarding AI safety?"

### Specify Domain and Sources

For high-quality results, mention source preferences:
- "According to peer-reviewed publications in high-impact journals..."
- "Based on FDA-approved treatments..."
- "From clinical trial registries like clinicaltrials.gov..."

### Structure Complex Queries

Break complex questions into clear components:
1. **Topic**: Main subject
2. **Scope**: Specific aspect of interest
3. **Context**: Time frame, domain, constraints
4. **Output**: Desired format or type of answer

**Example:**
"What improvements does AlphaFold3 offer over AlphaFold2 for protein structure prediction, according to research published between 2023 and 2024? Include specific accuracy metrics and benchmarks."

See `references/search_strategies.md` for comprehensive guidance on query design, domain-specific patterns, and advanced techniques.

## Common Use Cases

### Scientific Literature Search

python scripts/perplexity_search.py \ "What does recent research (2023-2024) say about the role of gut microbiome in Parkinson's disease? Focus on peer-reviewed studies and include specific bacterial species identified." \ --model sonar-pro

### Technical Documentation

python scripts/perplexity_search.py \ "How to implement real-time data streaming from Kafka to PostgreSQL using Python? Include considerations for handling backpressure and ensuring exactly-once semantics." \ --model sonar-reasoning-pro

### Comparative Analysis

python scripts/perplexity_search.py \ "Compare PyTorch versus TensorFlow for implementing transformer models in terms of ease of use, performance, and ecosystem support. Include benchmarks from recent studies." \ --model sonar-pro-search

### Clinical Research

python scripts/perplexity_search.py \ "What is the evidence for intermittent fasting in managing type 2 diabetes in adults? Focus on randomized controlled trials and report HbA1c changes and weight loss outcomes." \ --model sonar-pro

### Trend Analysis

python scripts/perplexity_search.py \ "What are the key trends in single-cell RNA sequencing technology over the past 5 years? Highlight improvements in throughput, cost, and resolution, with specific examples." \ --model sonar-pro

## Working with Results

### Programmatic Access

Use `perplexity_search.py` as a module:

from scripts.perplexity_search import search_with_perplexity

result = search_with_perplexity( query="What are the latest CRISPR developments?", model="openrouter/perplexity/sonar-pro", max_tokens=4000, temperature=0.2, verbose=False )

if result["success"]: print(result["answer"]) print(f"Tokens used: {result['usage']['total_tokens']}") else: print(f"Error: {result['error']}")

### Save and Process Results

Save to JSON

python scripts/perplexity_search.py "query" --output results.json

Process with jq

cat results.json | jq '.answer' cat results.json | jq '.usage'

### Batch Processing

Create a script for multiple queries:

#!/bin/bash queries=( "CRISPR developments 2024" "mRNA vaccine technology advances" "AlphaFold3 accuracy improvements" )

for query in "${queries[@]}"; do echo "Searching: $query" python scripts/perplexity_search.py "$query" --output "results_$(echo $query | tr ' ' '_').json" sleep 2 # Rate limiting done

## Cost Management

Perplexity models have different pricing tiers:

**Approximate costs per query:**
- Sonar: $0.001-0.002 (most cost-effective)
- Sonar Pro: $0.002-0.005 (recommended default)
- Sonar Reasoning Pro: $0.005-0.010
- Sonar Pro Search: $0.020-0.050+ (most comprehensive)

**Cost optimization strategies:**
1. Use `sonar` for simple fact lookups
2. Default to `sonar-pro` for most queries
3. Reserve `sonar-pro-search` for complex analysis
4. Set `--max-tokens` to limit response length
5. Monitor usage at https://openrouter.ai/activity
6. Set spending limits in OpenRouter dashboard