perplexity-reference-architecture

name	perplexity-reference-architecture
description	Implement Perplexity reference architecture with model routing, citation pipeline, and research automation. Use when designing new Perplexity integrations, reviewing project structure, or establishing architecture for search-augmented apps. Trigger with phrases like "perplexity architecture", "perplexity project structure", "how to organize perplexity", "perplexity design patterns".
allowed-tools	Read, Grep
version	1.12.0
license	MIT
author	Jeremy Longshore <[e-mail]>
tags	["saas","perplexity","perplexity-reference"]
compatibility	Designed for Claude Code

Perplexity Reference Architecture

Overview

Production architecture for AI-powered search with Perplexity Sonar API. Three tiers: search service (model routing + caching), citation pipeline (extract, validate, store), and research orchestrator (multi-query synthesis).

Architecture

┌─────────────────────────────────────────────┐
│              Application Layer              │
│  (Search Widget, Research Agent, Fact Check) │
└──────────────────────┬──────────────────────┘
                       │
┌──────────────────────▼──────────────────────┐
│            Search Service Layer             │
│  ┌──────────┐ ┌──────────┐ ┌─────────────┐ │
│  │  Model   │ │  Query   │ │   Response   │ │
│  │  Router  │ │  Cache   │ │   Parser     │ │
│  └──────────┘ └──────────┘ └─────────────┘ │
└──────────────────────┬──────────────────────┘
                       │
┌──────────────────────▼──────────────────────┐
│          api.perplexity.ai/chat/completions │
│  sonar | sonar-pro | sonar-reasoning-pro    │
└─────────────────────────────────────────────┘

Prerequisites

• Perplexity API key with Sonar access
• OpenAI-compatible client library (openai package)
• Redis for production caching (LRU for development)

Instructions

Step 1: Search Service with Model Routing

// src/perplexity/search-service.ts
import OpenAI from "openai";
import { createHash } from "crypto";

type SearchDepth = "quick" | "standard" | "deep" | "reasoning";

const MODEL_MAP: Record<SearchDepth, { model: string; maxTokens: number; timeout: number }> = {
  quick:     { model: "sonar",               : ,  :  },
  :  { : ,               : , :  },
  :      { : ,           : , :  },
  : { : , : , :  },
};

   {
  () {}

   () {
     config = [depth];
     cacheKey = .(query, config., opts);

    
     cached = ..(cacheKey);
     (cached && cached. > .()) {
       { ...cached., :  };
    }

     response =  ....({
      : config.,
      : [
        ...(opts. ? [{ :   , : opts. }] : []),
        { :   , : query },
      ],
      : config.,
      ...(opts. && { : opts. }),
      ...(opts. && { : opts. }),
    }  );

     result = {
      : response.[].. || ,
      : (response  ). || [],
      : (response  ). || [],
      : response.,
      : response.,
    };

    
     ttl = opts. ===  ?  : ;
    ..(cacheKey, { result, : .() + ttl });

     { ...result, :  };
  }

   (: , : , : ):  {
     ()
      .(.({ : query.().(), model, ...opts }))
      .();
  }
}