Perplexity

Perplexity is an AI search engine. For developers, the Sonar API provides grounded, cited answers for building search-enabled apps.

When to Use

• Real-Time Web-Augmented Search & Retrieval: Querying fresh, live web data with grounded, verifiable source citations.
• Fact-Checking & Research Automation: Generating comprehensive answers with linked URLs for financial, legal, and tech intelligence.
• Perplexity Sonar Models via API: Accessing sonar-pro, sonar, and reasoning models for production AI assistants.
• Recency-Filtered Search: Restricting search scope to the last day, week, or month for breaking developments.

Quick Start

from openai import OpenAI

# Perplexity uses OpenAI-compatible API format
client = OpenAI(
    api_key="your-perplexity-api-key",
    base_url="https://api.perplexity.ai"
)

response = client.chat.completions.create(
    model="sonar-pro",
    messages=[
        {"role": "user", "content": "What were the major tech announcements this week?"}
    ]
)

print(response.choices[0].message.content)

Core Concepts

Consuming Perplexity Sonar API with Citations

Querying live web search intelligence using OpenAI-compatible SDK:

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("PERPLEXITY_API_KEY"),
    base_url="https://api.perplexity.ai"
)

response = client.chat.completions.create(
    model="sonar-pro",
    messages=[
        {"role": "system", "content": "You are a research analyst. Provide factual answers with explicit citations."},
        {"role": "user", "content": "What are the latest developments in quantum computing error correction in 2026?"}
    ],
    temperature=0.2,
)

answer = response.choices[0].message.content
citations = getattr(response, 'citations', [])

print("Answer:\n", answer)
print("\nSources / Citations:")
for i, url in enumerate(citations, 1):
    print(f"[{i}] {url}")

Search Domain & Recency Filtering

Restricting search context to authoritative domains and specific timeframes:

import requests
import json

url = "https://api.perplexity.ai/chat/completions"
headers = {
    "Authorization": f"Bearer {os.environ.get('PERPLEXITY_API_KEY')}",
    "Content-Type": "application/json"
}

payload = {
    "model": "sonar",
    "messages": [
        {"role": "user", "content": "What were the key policy changes announced by the central bank this week?"}
    ],
    "search_domain_filter": ["bloomberg.com", "reuters.com", "ft.com"],
    "search_recency_filter": "week" # "day", "week", "month", "year"
}

res = requests.post(url, json=payload, headers=headers)
data = res.json()
print(data['choices'][0]['message']['content'])

Structured Output Formatting with Sonar

Directing the model to output verified markdown data tables:

response = client.chat.completions.create(
    model="sonar",
    messages=[
        {"role": "user", "content": "Compare the top 3 open-source vector databases by license, GitHub stars, and vector indexing algorithm. Output as a Markdown table."}
    ],
    temperature=0.1
)

print(response.choices[0].message.content)

Common Patterns

Domain-Restricted Search Grounding

Problem: Retrieval results include low-quality blogs or outdated forum answers.

Solution: Filter search sources using search_domain_filter:

response = client.chat.completions.create(
    model="sonar",
    messages=[{"role": "user", "content": "What is the latest RFC standard for HTTP/3?"}],
    extra_body={
        "search_domain_filter": ["ietf.org", "mozilla.org", "w3.org"],
        "return_citations": True
    }
)

print("Answer:", response.choices[0].message.content)
# Access citation URLs if enabled
citations = getattr(response, "citations", [])
print("Citations:", citations)

Best Practices

Do:

• Target sonar-pro for deep analytical research and sonar for fast, lightweight web queries.
• Inspect and parse the citations array to display clickable references in user-facing interfaces.
• Apply search_recency_filter when querying time-sensitive or breaking news topics.
• Use search_domain_filter to limit search to trusted corporate, scientific, or government websites.

Don't:

• Use high temperature settings (> 0.3) if exact factual precision and strict web grounding are desired.
• Assume web search is necessary for closed-domain math or pure coding tasks; standard LLMs are faster and cheaper.
• Hardcode PERPLEXITY_API_KEY in scripts; load securely from environment variables.

Troubleshooting

References

• Perplexity API