The Ultimate Guide to Perplexity Deep Research on macOS [With Examples]
Published on March 24, 2025   •   9 min read

The Ultimate Guide to Perplexity Deep Research on macOS [With Examples]

Jarek CeborskiJarek Ceborski

In the rapidly evolving landscape of AI technology, Deep Research represents a significant leap forward in how we interact with and extract value from large language models (LLMs). Today, we'll explore Perplexity's Sonar Deep Research model and how you can leverage its capabilities through Kerlig on your Mac.

What is Deep Research?

Deep Research is an advanced AI capability that goes beyond traditional LLM interactions. Unlike regular LLMs that provide answers based on their training data, Deep Research models actively search, browse, and synthesize information from multiple sources in real-time. They can:

  • Autonomously break down complex queries into manageable sub-tasks
  • Search and analyze hundreds of web pages
  • Cross-reference information from multiple sources
  • Synthesize findings into comprehensive, well-structured reports
  • Maintain context across long research sessions

What Makes Deep Research Different?

Traditional LLMs are like having a knowledgeable conversation partner with a fixed knowledge cutoff date. Deep Research models, on the other hand, are more like having a skilled research assistant who can:

  1. Actively Browse: They don't just search - they can navigate through websites, following relevant links and gathering context
  2. Think Iteratively: They plan their research approach, evaluate findings, and adjust their strategy as needed
  3. Synthesize Information: Instead of just summarizing, they can identify patterns, contradictions, and insights across multiple sources
  4. Maintain Long-Term Context: They can handle research sessions that span multiple interactions while keeping track of all gathered information

Technical Overview of Sonar Deep Research

Perplexity's Sonar Deep Research is a specialized model designed for multi-step retrieval, synthesis, and reasoning across complex topics. Built on a customized version of the open-source DeepSeek R1 model (as confirmed by Perplexity CEO Aravind Srinivas), it offers enhanced capabilities for comprehensive research and analysis. With a massive 200,000 token context window, it can process and analyze extensive amounts of information in a single session. The model excels at:

  • Autonomous searching and reading of sources
  • Continuous evaluation and refinement of research approach
  • Comprehensive report generation across various domains including: Finance, Marketing, Technology, Healthcare, Travel, Current Events, etc.

The DeepSeek R1 foundation provides Sonar Deep Research with robust capabilities for deep analysis and reasoning, while Perplexity's customizations enhance its research-specific features and integration with real-time web data.

Comparison: Traditional LLMs vs. Perplexity Sonar vs. Perplexity Sonar Deep Research

FeatureTraditional LLMsPerplexity SonarPerplexity Sonar Deep Research
Real-time Web Browsing❌ No - Limited to training data cutoff✅ Yes - Can search current web content✅ Yes - Advanced web crawling with link following and deep page analysis
Iterative Thinking❌ No - Single-pass responses✅ Limited - Basic follow-up capability✅ Advanced - Multi-step reasoning with strategy refinement
Information Synthesis✅ Basic - Single context summarization✅ Moderate - Multi-source integration✅ Advanced - Cross-reference validation and comprehensive synthesis
Context Maintenance✅ Limited - Usually single conversation✅ Moderate - Session memory✅ Extensive - 200K token context window with persistent memory
Complex Query Handling✅ Limited - Best for simple queries✅ Moderate - Can handle multi-part questions✅ Advanced - Breaks down and tackles complex research tasks
Source Verification❌ No - Based on training data✅ Basic - Links to sources✅ Advanced - Cross-references multiple sources with citations
Research Depth✅ Surface level✅ Moderate depth✅ In-depth analysis with multiple perspectives
Update Frequency❌ Fixed training data✅ Real-time web data✅ Real-time with historical context
Cost Efficiency✅ Low cost per query✅ Moderate cost✅ Usage-based pricing with optimization options
Response Time✅ Fast (seconds)✅ Quick (5-30 seconds)⚠️ Longer (1-5 minutes) but more comprehensive
Use CasesSimple queries, quick facts, general knowledgeModerate research, current events, basic analysisComprehensive research, detailed analysis, academic work

This expanded comparison highlights the key differences and capabilities of each model, helping you choose the right tool for your specific needs. While Traditional LLMs excel at quick responses and general knowledge, Perplexity Sonar adds real-time information access, and Sonar Deep Research provides the most comprehensive research capabilities with advanced reasoning and synthesis features.

When to Use Deep Research

Deep Research is particularly valuable when:

  • Comprehensive Information: You need detailed insights from multiple current sources, such as preparing a market analysis report or understanding the latest developments in technology.
  • Complex Relationships: Your query involves understanding intricate relationships between different topics, like analyzing the impact of economic policies on global markets.
  • Market Research or Competitive Analysis: You're evaluating competitors, their products, pricing strategies, customer feedback, and market positioning.
  • Rapidly Evolving Topics: You need to stay updated on fast-changing fields such as AI advancements, cryptocurrency trends, or emerging healthcare technologies.
  • Academic Literature Reviews: You're conducting thorough research for academic purposes, synthesizing findings from numerous scholarly articles and journals.
  • Trend Analysis: You need to analyze and predict trends across multiple sources, such as consumer behavior trends or industry growth forecasts.

Examples:

  • Conducting a detailed competitor analysis for a new software product launch.
  • Researching the latest advancements in renewable energy technologies.
  • Preparing an academic paper on recent developments in neuroscience.

When Not to Use Deep Research

Deep Research might be overkill for:

  • Simple Factual Queries: Quick lookups like historical dates, basic definitions, or straightforward facts.
  • Basic Calculations or Conversions: Tasks such as currency conversions, simple arithmetic, or unit conversions.
  • Creative Writing Tasks: Generating creative content like poems, stories, or imaginative scenarios.
  • Code Completion: Writing or debugging simple code snippets or scripts.
  • Quick Definitions or Explanations: Brief explanations or definitions that don't require extensive context or multiple sources.
  • Personal Opinions or Advice: Queries seeking subjective opinions, personal advice, or recommendations based on individual preferences.

Examples:

  • Finding the capital city of a country.
  • Calculating the conversion rate from USD to EUR.
  • Writing a short creative story or poem.
  • Debugging a simple Python script.
  • Getting a quick definition of a technical term.

Pricing Structure

Understanding the pricing model is important for efficient usage:

  1. Token Costs:

    • Input tokens: $2 per million tokens
    • Output tokens: $8 per million tokens
    • Input tokens include both prompt tokens and citation tokens from searches
  2. Search Operations:

    • Each search operation costs $5 per 1000 searches
    • Example: A request with 30 searches would cost $0.15
  3. Reasoning Process:

    • Distinct reasoning steps priced at $3 per million tokens
    • Covers extensive automated reasoning through research material
    • Separate from the Chain-of-Thought reasoning in the final output

This pricing structure reflects the model's sophisticated capabilities in conducting thorough research and analysis.

Cost Optimization Tips

To optimize your usage of Sonar Deep Research while managing costs effectively, focus on crafting specific and targeted queries that minimize unnecessary searches. Structure your research sessions by grouping related topics together to take advantage of the 200K token context window, which allows the model to build upon previous findings without repeating searches. When conducting research, break down complex tasks into focused questions and use the model's built-in citation feature to track sources efficiently. By maintaining focused research sessions and leveraging follow-up questions within the same context, you can avoid redundant searches and make the most of the model's memory capabilities. This approach not only reduces costs but also leads to more efficient and effective research outcomes.

Using Perplexity Sonar Deep Research in Kerlig: A Step-by-Step Guide

  1. Create a Perplexity API Key

  2. Add Your API Key to Kerlig

    • Go to Settings, Integrations tab and Select Perplexity
    • Paste your API key into the API Key field and click Save.
  3. Run Sonar Deep Research

    • Open Kerlig
    • Enter your prompt - anything you want to research
    • Select newly added Sonar Deep Research as model
    • Click Run or press Enter
    • Wait for the model to complete its research (this may take a few minutes depending on the complexity of your query)
    • Review the comprehensive report provided and ask follow-up questions if needed

Pro Tips for Better Results

  • Be specific in your queries
  • Break down complex research topics into focused questions
  • Use follow-up questions to dive deeper into specific aspects
  • Take advantage of Kerlig's document context features for more targeted research

Example Deep Research Prompts

Here are some well-structured prompts that work great with Deep Research:

1: Market Research for SaaS Product

Market Research for SaaS Product

Help me research the current state of AI-powered customer service solutions in the SaaS industry. Focus on:

  1. Leading companies and their products
  2. Key features and pricing strategies
  3. Customer satisfaction and reviews
  4. Market trends and growth projections
  5. Common challenges and solutions
Research results for SaaS product market

2: Travel Planning Research

Travel Planning Research

Research and create a comprehensive travel guide for Japan in Spring 2025:

  1. Best time to visit for cherry blossom viewing by region
  2. Top-rated ryokans and hotels with onsen (include pricing and booking windows)
  3. Transportation options between major cities
  4. Must-visit cultural festivals and events during spring
  5. Local food specialties and recommended restaurants
  6. Current travel restrictions and requirements Include recent traveler reviews and local tourism data.
Research results for travel guide for Japan

3: Quantum Computing Development

Quantum Computing Development

Research the latest developments in quantum computing from 2023-2025, specifically:

  1. Major breakthroughs in quantum error correction
  2. New quantum computing architectures
  3. Practical applications in cryptography
  4. Key research institutions and their findings
  5. Challenges and future directions Include citations from peer-reviewed sources.
Research results for recent quantum computing developments

4: Technology Trend Analysis

Technology Trend Analysis

Analyze the current state and future prospects of autonomous vehicles:

  1. Recent technological advancements
  2. Regulatory landscape across major markets
  3. Key players and their approaches
  4. Safety statistics and public perception
  5. Market predictions for 2024-2025 Focus on verified sources and industry reports.

5: Competitive Analysis

Competitive Analysis

Research the mobile payment industry in Southeast Asia:

  1. Market leaders and their market share
  2. User adoption rates by country
  3. Unique features and value propositions
  4. Integration with local banking systems
  5. Growth challenges and opportunities Include recent market data and analyst reports.

Tips for Writing Effective Deep Research Prompts:

  • Structure your query with clear sections or numbered points
  • Specify the time period for relevant information
  • Request specific types of sources (e.g., peer-reviewed, industry reports)
  • Include follow-up areas you want to explore
  • Be explicit about the depth of analysis needed

Conclusion

Deep Research models like Perplexity's Sonar represent a significant evolution in AI capabilities, offering a powerful tool for comprehensive research and analysis. Through Kerlig's intuitive macOS interface, you can now harness these capabilities alongside 350+ other AI models, making it an invaluable addition to your productivity toolkit.

Whether you're a researcher, professional, or simply someone who needs to stay well-informed, the combination of Perplexity's Deep Research capabilities and Kerlig's user-friendly interface provides a powerful solution for your information needs.


This article is part of the Kerlig Blog series, exploring the latest developments in AI technology and how they can enhance your productivity on macOS. Visit kerlig.com/blog for more insights and tutorials.