Perplexity Search API

What it is

The Perplexity Search API exposes the search infrastructure behind Perplexity’s public answer engine as a developer endpoint. It returns raw, ranked web results in real time, intended for grounding LLMs, powering agents, and feeding retrieval-augmented generation (RAG) pipelines. Perplexity describes the underlying index as spanning hundreds of billions of webpages.

The distinction that matters: this API returns source material, not a finished answer. You decide how to use it.

What sets it apart

• Snippet-level results. Instead of full documents, the API returns pre-ranked snippets — cutting much of the chunking and preprocessing a RAG pipeline would otherwise do.
• Freshness. The index is designed to update continuously, which reduces the risk of grounding a model on stale pages.
• Structured parsing. An internal parsing layer cleans unstructured web content into structured results before returning them.
• AI-oriented throughput. The infrastructure is tuned for high-volume, latency-sensitive AI workloads.

Search API vs. Sonar API

Perplexity ships two APIs for two jobs:

• Search API — returns raw, ranked results and snippets. Use it when you need to ground your own model, build a custom agent, or supply a RAG pipeline. It gives you the sources, not the conclusion.
• Sonar API — returns a synthesized, conversational answer, close to the experience on the Perplexity site. Use it when you want a ready-to-display response with no assembly on your side.

Reach for the Search API when you own the reasoning layer; reach for Sonar when you want the answer produced for you.

Benchmarking with search_evals

Perplexity has open-sourced search_evals, an evaluation framework for comparing search backends across single-step queries and multi-step agentic research. It lets you measure retrieval quality against alternatives on your own tasks rather than relying on vendor claims — which is the right way to evaluate any search backend before committing a pipeline to it.

Developer tooling

• Console for API-key management and monitoring.
• Documentation with guides and API references.
• Search SDK for faster prototyping and integration.

Where it fits

For agentic systems, fresh and accurate retrieval is a hard dependency. A snippet-level search API that returns clean, current, raw web data serves as the retrieval layer beneath agents and RAG pipelines — filling the gap left as several general-purpose search APIs have closed or been retired. Evaluate it the same way you would any dependency: benchmark it on your own workload, on the metrics that matter to your product.