Overview
Perplexity offers search-grounded LLMs – chat responses backed by live web search with inline citations, plus a longer-running research mode. The Automize connector covers Chat (sonar models) and Research (deep multi-step research) via the OpenAI-compatible Chat Completions endpoint.
- Regions: Single global API at api.perplexity.ai. Web search is global – responses include citation URLs that you should re-validate if you need a specific jurisdictional source.
- Plans: Pay-as-you-go. Pro / Enterprise tiers raise rate limits and add structured outputs + custom domains for citations.
Setting up the connection
Auth type: Bearer Token.
- In Perplexity → Settings → API → Generate API Key (starts with pplx-).
- Paste it as the API token. Connector sends Authorization: Bearer <key>.
Find your credentials at https://www.perplexity.ai/settings/api.
Rate limits
Per-minute rate limits vary by model – Sonar is more generous than Sonar Pro. Research mode is queue-backed (a single request can run for several minutes). Connector backs off on 429.
Available operations
2 operations available. Click any row to jump to its detail.
| Operation | Category | What it does |
|---|---|---|
| Chat Completion | General | Generate a web-grounded chat completion with citations |
| Deep Research | General | Run a Sonar deep-research query (longer-form, multi-step) |
Operations
Chat Completion
Generate a web-grounded chat completion with citations
| Parameter | Required | Type | Description |
|---|---|---|---|
Model model | No | text | |
Messages messages | No | text | |
Temperature temperature | No | text | |
Max Tokens max_tokens | No | text | |
Top P top_p | No | text | |
Search Domain Filter search_domain_filter | No | text | |
Search Recency Filter search_recency_filter | No | text | |
Return Citations return_citations | No | text | |
Return Images return_images | No | text |
Tips
- OpenAI-compatible Chat shape. model defaults to sonar; use sonar-pro for higher-quality answers + more aggressive search.
- search_recency_filter='day'|'week'|'month'|'year' bounds the citation window. Useful when you need recent-only sources.
- return_related_questions=true returns follow-up question suggestions alongside the answer.
- Citations come back in the response.citations[] array (URLs only). Cross-reference these for source attribution.
Deep Research
Run a Sonar deep-research query (longer-form, multi-step)
| Parameter | Required | Type | Description |
|---|---|---|---|
Model model | No | text | |
Messages messages | No | text | |
Temperature temperature | No | text | |
Max Tokens max_tokens | No | text | |
Return Citations return_citations | No | text |
Tips
- Multi-step research mode – fires multiple web searches and synthesises a long-form report. Slower (1-5 min) but cites 10-30 sources.
- Use for 'compile a brief on X' tasks where you'd otherwise stitch together several Chat calls.
- Async-ish: the connector waits on the call but the upstream call can take several minutes – set a generous timeout on the calling step.
FAQ
- Sonar vs Sonar Pro – when do I pay for Pro?
- Sonar is fine for short factual lookups (one search, one answer). Sonar Pro runs more searches per question and gives longer, better-cited answers – worth it for research summaries, comparison questions, recency-sensitive queries.
- How current are the citations?
- Real-time search at the moment of the call. If you need a specific cutoff, set search_recency_filter and don't cache responses – Perplexity reaches out to the web each time.
- Function calling / tools?
- Not supported on Sonar models – they're search-grounded, not tool-using. For tool-call workflows, route to OpenAI / Anthropic / Mistral and use Perplexity only for the web-grounded sub-step.
- JSON mode?
- Yes – response_format={type:'json_schema', json_schema:{...}}. Sonar Pro only; Sonar returns prose. Structured outputs are useful when piping the answer + citations into a downstream pipeline.