perplexity-cache

A tiny TypeScript wrapper around the official Perplexity SDK that adds transparent caching for:

• client.search.create(...)
• client.responses.create(...)

It is designed as a drop-in API-compatible surface so you can keep your existing calling code and reduce duplicate API calls during iteration.

• Avoid paying repeatedly for identical prompts during development.
• Speed up repeated queries with local cache hits.
• Keep the same method shape as the original Perplexity client (search.create and responses.create).

PerplexityCached wraps a regular Perplexity client and a Cacheable instance.

• On create(...), it builds a deterministic SHA-256 cache key from args.
• If cache is enabled and a value exists, it returns the cached value.
• Otherwise, it calls the network API and stores the response.

• search.create: key = SHA-256 of normalized args.
  • If query is an array, it is normalized to a newline-joined string before hashing.
• responses.create: key = SHA-256 of responses:${JSON.stringify(args)}.

This keeps search and responses keyspaces independent.

• Node.js 18+
• A valid Perplexity API key in PERPLEXITY_API_KEY
• Dependencies:
  • @perplexity-ai/perplexity_ai
  • cacheable
  • a cache store implementation (examples use @keyv/sqlite)

From this repository root:

npm install @perplexity-ai/perplexity_ai cacheable @keyv/sqlite

If you run examples with tsx, install it where you execute commands:

npm install -D tsx typescript

import Path from "node:path";
import KeyvSqlite from "@keyv/sqlite";
import { Cacheable, KeyvStoreAdapter } from "cacheable";
import { Perplexity } from "@perplexity-ai/perplexity_ai";
import { PerplexityCached } from "./src/perplexity_cached";

// Create the base Perplexity client with your API key
const baseClient = new Perplexity({
	apiKey: process.env.PERPLEXITY_API_KEY!,
});

// Create a Cacheable instance with a SQLite store - you can use any Keyv-compatible store or even an in-memory one for testing
const sqlitePath = Path.resolve(__dirname, ".perplexity_cache.sqlite");
const store = new KeyvSqlite(`sqlite://${sqlitePath}`);
const cacheable = new Cacheable({ secondary: store as KeyvStoreAdapter });

// Create a PerplexityCached instance that wraps the base client and cacheable
const perplexityCached = new PerplexityCached(baseClient, cacheable);

// Now you can use perplexityCached just like the original client, but with caching
const response = await perplexityCached.responses.create({
	model: "perplexity/sonar",
	input: "Summarize this page: https://example.com",
	tools: [{ type: "fetch_url" }],
});

// display the result
console.log(response.output_text);

new PerplexityCached(perplexityClient, cache, { cacheEnabled?: boolean })

• perplexityClient: instance of Perplexity
• cache: instance of Cacheable
• cacheEnabled (optional, default true):
  • true: read from cache and fallback to network on miss
  • false: skip cache reads, always call network, still write results to cache

The wrapper exposes:

• client.search.create(args)
• client.responses.create(args)
• client.setCacheEnabled(boolean)
• await client.cleanCache()

Both methods keep the input/output typing from the official SDK.

From repository root:

# search.create with visible cache-hit timing
npx tsx ./examples/search_create.ts

# responses.create with fetch_url tool
npx tsx ./examples/response_create.ts

# pro-search preset response example
npx tsx ./examples/pro_search.ts

• No TTL is enforced by this wrapper directly. Configure expiration in your Cacheable setup if needed.
• Cached response payloads are returned as-is.
• Cache key stability depends on argument structure and ordering in provided objects.

perplexity-cache/
  src/perplexity_cached.ts
  examples/
    search_create.ts
    response_create.ts
    pro_search.ts