Perplexity Data Handling

Overview

Manage search query data and results from Perplexity Sonar API. Covers query sanitization, citation validation, result caching with freshness policies, and conversation context management for research workflows.

Prerequisites

• Perplexity API key
• OpenAI-compatible client library
• Understanding of search result data structures
• Cache storage for results

Instructions

Step 1: Query Sanitization

function sanitizeQuery(query: string): string {
  // Remove PII that might leak into search queries
  let clean = query
    .replace(/\b[\w.+-]+@[\w-]+\.[\w.]+\b/g, '[email]')
    .replace(/\b\d{3}[-.]?\d{3}[-.]?\d{4}\b/g, '[phone]')
    .replace(/\b\d{3}-\d{2}-\d{4}\b/g, '[ssn]');

  // Remove overly specific identifiers
  clean = clean
    .replace(/\b(user|customer|account)\s*#?\s*\d+\b/gi, '[ID]')
    .replace(/\b[A-Z0-9]{20,}\b/g, '[TOKEN]');

  return clean;
}

async function safeSearch(rawQuery: string) {
  const query = sanitizeQuery(rawQuery);

  const result = await perplexity.chat.completions.create({
    model: 'sonar',
    messages: [{ role: 'user', content: query }],
  });

  return result;
}

Step 2: Citation Validation and Cleaning

interface ValidatedCitation {
  url: string;
  domain: string;
  isAccessible: boolean;
  title?: string;
}

function extractAndValidateCitations(responseText: string): ValidatedCitation[] {
  const urlRegex = /https?:\/\/[^\s\])"]+/g;
  const urls = [...new Set(responseText.match(urlRegex) || [])];

  return urls.map(url => {
    try {
      const parsed = new URL(url);
      return {
        url: url.replace(/[.,;:]+$/, ''), // Clean trailing punctuation
        domain: parsed.hostname,
        isAccessible: true,
      };
    } catch {
      return { url, domain: 'unknown', isAccessible: false };
    }
  }).filter(c => c.isAccessible);
}

function deduplicateCitations(citations: ValidatedCitation[]): ValidatedCitation[] {
  const seen = new Set<string>();
  return citations.filter(c => {
    const key = c.domain + c.url.split('?')[0]; // Ignore query params
    if (seen.has(key)) return false;
    seen.add(key);
    return true;
  });
}

Step 3: Result Caching with Freshness Policy

import { LRUCache } from 'lru-cache';
import { createHash } from 'crypto';

interface CachedResult {
  response: string;
  citations: ValidatedCitation[];
  cachedAt: number;
  queryHash: string;
}

// Different TTLs based on query type
const CACHE_TTL = {
  factual: 1000 * 60 * 60 * 24,  // 24 hours for stable facts  # 1000: 1 second in ms
  news: 1000 * 60 * 30,           // 30 min for news queries  # 1 second in ms
  research: 1000 * 60 * 60 * 4,   // 4 hours for research  # 1 second in ms
  default: 1000 * 60 * 60,        // 1 hour default  # 1 second in ms
};

const resultCache = new LRUCache<string, CachedResult>({ max: 500 });  # HTTP 500 Internal Server Error

function detectQueryType(query: string): keyof typeof CACHE_TTL {
  if (/\b(latest|today|breaking|recent)\b/i.test(query)) return 'news';
  if (/\b(research|study|paper|analysis)\b/i.test(query)) return 'research';
  if (/\b(what is|define|how does)\b/i.test(query)) return 'factual';
  return 'default';
}

async function cachedSearch(query: string) {
  const hash = createHash('sha256').update(query.toLowerCase().trim()).digest('hex');
  const cached = resultCache.get(hash);
  if (cached) return cached;

  const result = await safeSearch(query);
  const content = result.choices[0].message.content || '';
  const citations = deduplicateCitations(extractAndValidateCitations(content));
  const queryType = detectQueryType(query);

  const entry: CachedResult = {
    response: content,
    citations,
    cachedAt: Date.now(),
    queryHash: hash,
  };

  resultCache.set(hash, entry, { ttl: CACHE_TTL[queryType] });
  return entry;
}

Step 4: Conversation Context Limits

class ResearchContext {
  private messages: any[] = [];
  private maxMessages = 10;
  private maxTokenEstimate = 8000;  # 8000: API server port

  addMessage(role: string, content: string) {
    this.messages.push({ role, content });

    // Trim oldest messages if over limit
    while (this.messages.length > this.maxMessages) {
      this.messages.shift();
    }

    // Trim if estimated tokens too high
    while (this.estimateTokens() > this.maxTokenEstimate && this.messages.length > 2) {
      this.messages.splice(1, 1); // Remove second oldest (keep system prompt)
    }
  }

  getMessages() { return [...this.messages]; }
  clear() { this.messages = []; }

  private estimateTokens(): number {
    return this.messages.reduce((sum, m) => sum + Math.ceil(m.content.length / 4), 0);
  }
}

Error Handling

Issue	Cause	Solution
PII in search query	User entered personal data	Apply sanitizeQuery before API call
Broken citations	URL changed or removed	Validate URLs, remove inaccessible ones
Stale cached results	TTL too long for news queries	Use query-type-aware TTL
Context overflow	Too many conversation turns	Trim old messages automatically

Examples

Research Session with Caching

const context = new ResearchContext();
context.addMessage('system', 'You are a research assistant.');

const result = await cachedSearch('Latest advances in quantum computing 2025');  # 2025 year
console.log(`Response: ${result.response.slice(0, 200)}...`);  # HTTP 200 OK
console.log(`Citations: ${result.citations.length} sources`);

Resources

• Perplexity API Docs
• Perplexity Data Policy

Output

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