Skip to main content

​
Overview

The Perplexity Sonar Pro Agent provides real-time, source-backed information from the live web. It specializes in research, news, competitive analysis, and fact verification.

​
Initialization

# agent/agent_factory.py:234-245
perplexity_agent = Agent(
    id="pplx-agent",
    name="Perplexity Sonar Pro",
    model=OpenAILike(
        id="sonar-pro",
        base_url=PROVIDER,
        api_key=CUSTOM_PROVIDER_API_KEY
    ),
    add_datetime_to_context=True,
    timezone_identifier="Asia/Kolkata",
)

​
Model Configuration

Uses Perplexity’s Sonar Pro model through OpenAI-compatible API: 
model=OpenAILike(
    id="sonar-pro",
    base_url=PROVIDER,
    api_key=CUSTOM_PROVIDER_API_KEY
)

​
Model Capabilities

  • Real-time web access: Searches current web content
  • Source attribution: Returns sources for all claims
  • Contextual understanding: Uses current date/time for relevance
  • Multi-source verification: Cross-references information

​
Use Cases

The Perplexity Agent is delegated tasks involving:

​
1. Deep Research

From system_prompt.md:79:
Deep research / real-time web data / complex analysis → delegate to pplx-agent.
  • Comprehensive topic analysis
  • Multi-source research
  • Competitive analysis
  • Detailed investigations

​
2. Real-Time Information

  • Current events and news
  • Market data and trends
  • Live updates
  • Breaking information

​
3. Fact Verification

From system_prompt.md:93-99: 
## Accuracy, verification & citations (CRITICAL)

* **Always verify facts**, statistics, time-sensitive claims, and numbers using web/search tools or data connectors before presenting them as truth.
* Cross-check high-impact claims with at least two reputable sources.
* Cite sources succinctly (one-line attribution or clickable link if supported). Use credibility indicators (site reputation, publication date) when relevant.
* If information cannot be verified, state uncertainty: "I couldn't verify X; here's what I found…".
The agent excels at:
  • Verifying statistics and numbers
  • Cross-checking claims
  • Providing source citations
  • Assessing information credibility

​
4. Complex Analysis

  • Comparative studies
  • Trend analysis
  • Expert opinion gathering
  • Technical research

​
Tools

The Perplexity Agent does not have explicit tools - its capabilities come from the Sonar Pro model’s native web access: 
tools=None  # Implicit - no tools defined
The model inherently:
  • Searches the web
  • Retrieves sources
  • Analyzes content
  • Generates citations

​
Context Awareness

add_datetime_to_context=True,
timezone_identifier="Asia/Kolkata",
The agent receives:
  • Current date and time (IST)
  • Message timestamps
  • Temporal context for queries
This enables:
  • Time-sensitive searches (“today’s news”)
  • Date-aware research (“recent developments”)
  • Historical context (“what happened yesterday”)

​
Configuration

Environment variables: 
# Provider endpoint (must support Perplexity models)
PROVIDER=https://your-openai-compatible-endpoint

# API key with Perplexity access
CUSTOM_PROVIDER_API_KEY=your_api_key

​
Delegation Strategy

​
When to Use Perplexity Agent

✅ Use for:
  • Web research and information gathering
  • Current events and news
  • Fact verification
  • Source-backed answers
  • Competitive analysis
❌ Do NOT use for:
  • Code execution (use Code Agent or Groq Compound)
  • Chat history analysis (use Context Q&A Agent)
  • Math calculations (use Team Leader’s CalculatorTools)

​
Fallback Behavior

From system_prompt.md:87-89:
If the chosen agent is unavailable or fails, attempt one fallback (next appropriate agent) before returning a best-effort partial answer.
If Perplexity Agent fails:
  1. Try Code Agent with ExaTools for web search
  2. Return best-effort answer with uncertainty disclaimer

​
Response Guidelines

The agent follows these citation standards: 
* Cite sources succinctly (one-line attribution or clickable link if supported)
* Use credibility indicators (site reputation, publication date) when relevant
* If information cannot be verified, state uncertainty: "I couldn't verify X; here's what I found…"

​
Example Queries

Good delegation to Perplexity:
  • “What are the latest developments in AI regulation?”
  • “Compare the top 5 CRM platforms for startups”
  • “What’s the current status of the XYZ merger?”
  • “Find research papers on quantum computing from 2024”
Better handled by other agents:
  • “Calculate the ROI for this investment” → Team Leader (CalculatorTools)
  • “Run this Python script” → Code Agent or Groq Compound
  • “What did I say about this topic last week?” → Context Q&A Agent

​
Performance Characteristics

  • Speed: Moderate (web search adds latency)
  • Accuracy: High (source-backed)
  • Cost: Higher (Sonar Pro is premium)
  • Token usage: Variable (depends on sources)

​
Best Practices

  1. Clear queries: Specific questions get better results
  2. Time bounds: Specify date ranges when relevant
  3. Source requirements: Request specific source types if needed
  4. Verification level: Indicate if high-confidence sources are required

​
Next Steps

Build docs developers (and LLMs) love

Get started for freeTalk to us