The Perplexity Search API (www.perplexity.ai)

Perplexity has launched a Search API that lets developers embed conversational, web-backed answers into apps by combining on‑the‑fly web retrieval with large‑language models. Instead of returning a list of links, the API returns concise, natural‑language answers accompanied by inline citations and source links; it supports streaming responses for low latency, configurable retrieval depth, and developer tooling (REST endpoints and SDKs). Under the hood the service uses retrieval‑augmented generation (RAG): a real‑time search layer pulls relevant documents from the web, then one or more LLMs synthesize an answer while surfacing provenance to reduce hallucinations. This matters because it bridges traditional search and generative AI, making grounded, citation‑aware Q&A easy to integrate into assistants, research tools, and knowledge workflows. For ML practitioners the key technical implications are obvious: using RAG reduces factual drift but introduces complexity around indexing, freshness, and source reliability; streaming and hybrid model routing are crucial to trade off latency vs. accuracy; and there are practical concerns about rate limits, cost, and content licensing when crawling or citing third‑party pages. The API accelerates production use of grounded LLMs but also raises questions about attribution, legal compliance, and how citation quality will be measured and enforced.