Perplexity Search API on LLM Gateway

An agent that answers questions about current events needs fresh web results, and most of the time it needs the results themselves, not another model's summary of them. Until now, the only way to reach Perplexity through LLM Gateway was Sonar, which searches and then writes prose. The Perplexity Search API is now available at /v1/search: ranked web results with extracted page content, on the same key, bill and logs as every model you already call.

Raw search results or a search-grounded answer

Both approaches exist on the gateway, and they solve different problems.

Native web search is the shortest path to an answer, but the model decides what to search and what to keep. The Search API hands that decision back to you: you get the URLs and snippets, and you choose what goes into the context window.

Call the Perplexity Search API

The endpoint mirrors Perplexity's request and response, so existing Perplexity Search code only needs a new base URL and key.

1curl https://api.llmgateway.io/v1/search \2  -H "Authorization: Bearer $LLM_GATEWAY_API_KEY" \3  -H "Content-Type: application/json" \4  -d '{5    "query": "latest developments in open-source LLMs",6    "max_results": 5,7    "search_recency_filter": "week",8    "search_domain_filter": ["arxiv.org", "huggingface.co"]9  }'

1curl https://api.llmgateway.io/v1/search \2  -H "Authorization: Bearer $LLM_GATEWAY_API_KEY" \3  -H "Content-Type: application/json" \4  -d '{5    "query": "latest developments in open-source LLMs",6    "max_results": 5,7    "search_recency_filter": "week",8    "search_domain_filter": ["arxiv.org", "huggingface.co"]9  }'

1{2  "id": "9ed15dce-a498-40b2-8bc9-7f13f35901cd",3  "model": "perplexity/perplexity-search",4  "results": [5    {6      "title": "…",7      "url": "https://…",8      "snippet": "…",9      "date": "2026-09-25",10      "last_updated": "2026-09-27"11    }12  ],13  "server_time": null14}

1{2  "id": "9ed15dce-a498-40b2-8bc9-7f13f35901cd",3  "model": "perplexity/perplexity-search",4  "results": [5    {6      "title": "…",7      "url": "https://…",8      "snippet": "…",9      "date": "2026-09-25",10      "last_updated": "2026-09-27"11    }12  ],13  "server_time": null14}

Every Perplexity filter is forwarded unchanged: country, language, domain allow and deny lists, recency, publication and last-updated date ranges, and per-result or total token budgets for extracted content. Pass query as an array of up to five related queries to run them in a single request.

Give any model a search tool

Because the results are plain JSON, the Search API works as a tool for any model on the gateway. The handler below runs a search and returns compact results for the model to read and cite.

1async function searchWeb(query: string) {2  const res = await fetch("https://api.llmgateway.io/v1/search", {3    method: "POST",4    headers: {5      Authorization: `Bearer ${process.env.LLM_GATEWAY_API_KEY}`,6      "Content-Type": "application/json",7    },8    body: JSON.stringify({9      query,10      search_type: "fast",11      max_results: 5,12    }),13  });14  const { results } = await res.json();15  return results.map((r: { title: string; url: string; snippet: string }) => ({16    title: r.title,17    url: r.url,18    snippet: r.snippet,19  }));20}

1async function searchWeb(query: string) {2  const res = await fetch("https://api.llmgateway.io/v1/search", {3    method: "POST",4    headers: {5      Authorization: `Bearer ${process.env.LLM_GATEWAY_API_KEY}`,6      "Content-Type": "application/json",7    },8    body: JSON.stringify({9      query,10      search_type: "fast",11      max_results: 5,12    }),13  });14  const { results } = await res.json();15  return results.map((r: { title: string; url: string; snippet: string }) => ({16    title: r.title,17    url: r.url,18    snippet: r.snippet,19  }));20}

Register searchWeb as a function tool in the tools array of a /v1/chat/completions request and return its output as the tool result. The search call and the model call both land in the same activity log, so you can see what each answer was built from.

Pricing: two models, billed per search

model is optional. Leave it out and Perplexity's search_type selects the model; set it explicitly to pin one.

Billing follows Perplexity's own rules:

• No token charges, however much page content a search returns.
• One charge per request, even when query holds several queries.
• Failed requests are free. An upstream error is logged with a cost of zero.

Search requests count toward the same credits, spend limits, IAM rules and data retention settings as the rest of your traffic. Perplexity's people search type is not supported yet and returns a 400.

Get started

• Try LLM Gateway free
• Read the Search API docs
• Compare with native web search