Perplexity Architecture Variants

Overview

Deployment architectures for Perplexity Sonar search API at different scales. Perplexity's search-augmented generation model fits different patterns from simple search widgets to full research automation pipelines.

Prerequisites

• Perplexity API key configured
• Clear search/research use case
• Infrastructure for chosen scale

Instructions

Step 1: Direct Search Widget (Simple)

Best for: Adding AI search to an app, < 500 queries/day.

@app.route('/ask')
def ask():
    response = pplx_client.chat.completions.create(
        model="sonar", messages=[{"role": "user", "content": request.args["q"]}]
    )
    return jsonify({
        "answer": response.choices[0].message.content,
        "citations": response.citations
    })

Step 2: Cached Research Layer (Moderate)

Best for: Repeated queries, 500-5K queries/day, research tools.

class CachedResearch:
    def __init__(self, client, cache, ttl=1800):  # 1800: timeout: 30 minutes
        self.client = client
        self.cache = cache
        self.ttl = ttl

    def search(self, query: str, model: str = "sonar"):
        key = f"pplx:{hashlib.sha256(query.encode()).hexdigest()}"
        cached = self.cache.get(key)
        if cached:
            return json.loads(cached)
        result = self.client.chat.completions.create(
            model=model, messages=[{"role": "user", "content": query}]
        )
        data = {"answer": result.choices[0].message.content, "citations": result.citations}
        self.cache.setex(key, self.ttl, json.dumps(data))
        return data

Step 3: Multi-Query Research Pipeline (Scale)

Best for: Automated research, 5K+ queries/day, report generation.

class ResearchPipeline:
    async def research_topic(self, topic: str) -> dict:
        # Decompose into sub-questions
        sub_questions = await self.decompose(topic)
        # Run parallel searches
        results = await asyncio.gather(*[
            self.search_with_cache(q) for q in sub_questions
        ])
        # Synthesize into report
        report = await self.synthesize(topic, results)
        return {"topic": topic, "sections": results, "synthesis": report}

    async def decompose(self, topic: str) -> list[str]:
        r = self.client.chat.completions.create(
            model="sonar", messages=[
                {"role": "system", "content": "Break this topic into 3-5 specific research questions."},
                {"role": "user", "content": topic}
            ])
        return r.choices[0].message.content.strip().split("\n")

Decision Matrix

Error Handling

Resources

• Perplexity API Docs

Output

• Configuration files or code changes applied to the project
• Validation report confirming correct implementation
• Summary of changes made and their rationale

Examples

Basic usage: Apply perplexity architecture variants to a standard project setup with default configuration options.

Advanced scenario: Customize perplexity architecture variants for production environments with multiple constraints and team-specific requirements.