Perplexity

Perplexity exposes the Sonar line of search-augmented models through its API platform, alongside Search and Embeddings APIs. The chat surface is OpenAI-compatible at https://api.perplexity.ai for older Sonar models, and the Agent API at https://api.perplexity.ai/v1/agent exposes a Responses-style surface that fronts frontier third-party models with web search tools.

Create the Client (Sonar Chat)

import { AgentBuilder } from "@anvia/core";
import { OpenAIClient } from "@anvia/openai";

const client = new OpenAIClient({
  baseUrl: "https://api.perplexity.ai",
  apiKey: process.env.PERPLEXITY_API_KEY,
});

const model = client.completionModel("sonar-pro");

const agent = new AgentBuilder("research", model)
  .instructions("Answer with citations.")
  .build();

const response = await agent.prompt("Latest news about renewable energy policy.").send();

console.log(response.output);

baseUrl makes Anvia use the OpenAI-compatible chat completion adapter. The model id is the Sonar id, not an Anvia-specific alias.

Agent API (Frontier Models)

The Agent API fronts models such as openai/gpt-5.5 and openai/gpt-5.2 with built-in web search, X search, and source filtering. It uses a Responses-style shape rather than the chat completions surface, so wire it up directly with fetch or a Perplexity SDK when you need it.

const response = await fetch("https://api.perplexity.ai/v1/agent", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.PERPLEXITY_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    model: "openai/gpt-5.5",
    input: "Today's most important AI announcements",
    tools: [{ type: "web_search" }],
  }),
});

const data = await response.json();
console.log(data.output_text);

Use the Agent API when you want the search-augmented frontier models and the source-filter controls. Use the Sonar chat surface when you want a single chat-completion model that Perplexity runs directly.

Get the Model List

Perplexity does not expose a generic /v1/models endpoint. Pick a model id from the Perplexity model catalog and pass it to completionModel(...). Sonar family ids include sonar, sonar-pro, sonar-reasoning, and sonar-reasoning-pro.

Notes

• Perplexity API keys are passed as bearer tokens in the Authorization header.
• The Agent API supports web_search tools with search_domain_filter and search_recency_filter controls. JSON schema structured outputs are supported through response_format.
• Sonar chat models return citations in the response. The citations surface differs from the Agent API; downstream code that expects OpenAI-shaped responses may need a small adapter.
• Perplexity also publishes a Search API for raw, ranked web results, and an Embeddings API for retrieval workloads. Use those for non-chat workloads.

For current Perplexity API details, see the Perplexity API docs and the Sonar model cards.