Get a single grounded, cited answer to a natural-language question using Perplexity's Sonar chat completions endpoint.

Documented steps

1. Get an API key from the Perplexity API settings and store it as PERPLEXITY_API_KEY; every request needs headers Authorization: Bearer $PERPLEXITY_API_KEY and Content-Type: application/json (https://docs.perplexity.ai/docs/getting-started/quickstart).
2. POST to https://api.perplexity.ai/chat/completions (an OpenAI-compatible alias) — the docs state the canonical path is POST /v1/sonar, with /chat/completions accepted for SDK compatibility (https://docs.perplexity.ai/docs/sonar/openai-compatibility).
3. Body: {"model": "sonar", "messages": [{"role": "user", "content": "<question>"}]} — model is required and is one of the enum values sonar, sonar-pro, sonar-reasoning-pro, sonar-deep-research (https://docs.perplexity.ai/api-reference/sonar-post).
4. Read the answer text from choices[0].message.content — same shape as OpenAI chat completions (id, model, created, choices, usage, object: "chat.completion").
5. Read grounding data from the top-level citations array (URL strings) and search_results array (objects with source metadata) rather than trusting URLs the model may write inline in the text (https://docs.perplexity.ai/api-reference/sonar-post).
6. Check usage for prompt_tokens/completion_tokens/total_tokens plus Perplexity's cost breakdown before scaling call volume.