Perplexity

Search the web and chat with Perplexity AI

The Perplexity server provides access to Perplexity AI’s search and chat capabilities, enabling web searches with recency filters and conversations with various Perplexity models.

Authentication

This server uses API key authentication.

Setup

1. Obtain a Perplexity API key from Perplexity AI
2. Run the authentication setup:

python src/servers/perplexity/main.py auth

3. Enter your API key when prompted

The API key is stored securely in your local authentication configuration.

Models

The server supports these Perplexity models:

Model	Description
sonar	Standard search model (default)
sonar-pro	Enhanced search with better accuracy
sonar-deep-research	In-depth research mode
sonar-reasoning	Reasoning-focused responses
sonar-reasoning-pro	Advanced reasoning capabilities

Prompts

The server provides pre-configured prompts for common use cases.

search_with_recency

Search with time-based filtering.

Arguments:

• query (required): The search query
• recency (optional): Time filter - hour, day, week, month, year, or none

Example Usage:

Query: "latest AI developments"
Recency: week

Generates a prompt optimized for time-filtered web searches.

code_assistant

Get coding help from Perplexity.

Arguments:

• problem (required): The coding problem or question
• language (optional): Programming language

Example Usage:

Problem: "How to implement binary search"
Language: Python

Generates a prompt optimized for code-related queries.

Tools

search

Search the web using Perplexity’s API with optional filters.

Parameters:

• query (string, required): The search query
• model (string, optional): Model to use (default: sonar)
• recency_filter (string, optional): Time filter - hour, day, week, month, year
• return_related (boolean, optional): Include related questions

Example:

{
  "query": "best practices for API design",
  "model": "sonar-pro",
  "recency_filter": "month",
  "return_related": true
}

Search Context: The search uses high search context size for comprehensive results.

Response Format:

• Main answer content
• Related questions (if requested)

Related Questions Example:

**Related Questions:**
- What are RESTful API design principles?
- How to version APIs effectively?
- What is GraphQL and when to use it?

chat

Send a message to a Perplexity model without forcing web search.

Parameters:

• message (string, required): The message to send
• model (string, optional): Model to use (default: sonar)
• system_prompt (string, optional): Custom system prompt (default: “Be helpful, concise, and precise.”)
• temperature (number, optional): Randomness of responses, 0-2 (default: 0.7)

Example:

{
  "message": "Explain how neural networks work",
  "model": "sonar-reasoning",
  "system_prompt": "You are an expert AI educator. Explain concepts clearly with examples.",
  "temperature": 0.5
}

Search Context: Uses medium search context size for balanced performance.

Temperature Guide:

• 0.0-0.3: More focused and deterministic
• 0.4-0.7: Balanced creativity and consistency
• 0.8-2.0: More creative and varied responses

Recency Filters

Time-based filtering for search results:

Filter	Description
hour	Results from the past hour
day	Results from the past 24 hours
week	Results from the past 7 days
month	Results from the past 30 days
year	Results from the past year
none	No time filtering (default)

API Reference

Base URL: https://api.perplexity.ai

Key Endpoint

POST /chat/completions

Request format:

{
  "model": "sonar",
  "messages": [
    {"role": "system", "content": "System prompt"},
    {"role": "user", "content": "User message"}
  ],
  "temperature": 0.7,
  "return_related_questions": false,
  "web_search_options": {
    "search_context_size": "high"
  },
  "search_recency_filter": "week"
}

Response format:

{
  "choices": [
    {
      "message": {
        "content": "Response text"
      }
    }
  ],
  "related_questions": ["Question 1", "Question 2"]
}

Features

Web Search

• Configurable search context size (medium/high)
• Time-based filtering for recent information
• Related questions for exploration
• Model selection for different use cases

Chat

• Flexible system prompts
• Temperature control for response variety
• Multiple specialized models
• Medium search context for background information

Error Handling

The server handles:

• Missing API key errors
• Invalid model names (validates against supported models)
• API rate limits and timeouts (60 second timeout)
• HTTP error responses with status codes
• Network failures

Errors are returned with descriptive messages including:

• Status codes from Perplexity API
• Error text from responses
• Exception details for debugging

Best Practices

1. Model Selection:

   • Use sonar for general searches
   • Use sonar-pro for higher accuracy needs
   • Use sonar-deep-research for comprehensive research
   • Use reasoning models for complex problem-solving

2. Recency Filters:

   • Apply time filters for news and current events
   • Use none for timeless topics
   • Consider week or month as balanced defaults

3. Temperature:

   • Lower (0.0-0.3) for factual, consistent responses
   • Medium (0.4-0.7) for balanced output
   • Higher (0.8-2.0) for creative tasks

4. Related Questions:

   • Enable for exploratory research
   • Use to discover related topics
   • Helpful for user guidance