Perplexity
AIGet web-grounded AI answers with Perplexity Sonar, run the standalone Search API, use the Agent API with built-in research tools, and create embeddings, all with live citations.
What This Integration Enables
Perplexity is built for grounded answers. Sonar generates responses from live web sources and returns citations, the Search API returns ranked results, the Agent API runs multi-step research with built-in tools, and embeddings support retrieval over your own content. The citations are the point: an answer you can trace to a source is one a person can verify. Web-grounded and cited beats plausible-but-unsourced for any answer that will be acted on. But a cited answer is still a starting point a person should confirm. An orchestration layer runs the research, checks the sources, and routes the brief to a human before it is used. FlowRunner is built for that layer, so research becomes a supervised step rather than an unchecked auto-answer.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Grounded Market Briefs
An analyst requests current context on a market or company. The agent asks Sonar for a grounded answer with citations, verifies the sources resolve and are recent, and assembles a sourced brief. The analyst gets a draft with every citation linked and verifies it before use. Research that meant opening a dozen tabs becomes a checkable, sourced draft.
Answering With Live Data
A workflow needs answers that reflect current information, not a model's training cutoff. The agent uses Sonar to answer from live sources with citations. Well-sourced answers return directly for internal use; anything customer-facing or where the sources conflict is routed to a person to confirm before it goes out.
Multi-Step Research
A question needs several steps of research to answer well. The agent uses the Agent API's built-in research tools to work through it and returns a sourced conclusion. A person reviews the reasoning and the sources before the conclusion is acted on, so the automated research is fast and still accountable.
Human-in-Loop Highlight
A cited answer is easier to trust and still not something to act on unverified. When Perplexity produces a brief or an answer that will drive a decision or reach a customer, FlowRunner routes it through a [human-in-the-loop](/concepts/human-in-the-loop/) step: the agent pauses, presents the answer with every source link, and sends it to the analyst via Slack. They verify the sources and confirm. Grounded research does the gathering; a person owns the conclusion.
Agent Capabilities
11 actionsSonar
2- Ask Asks Perplexity a question and returns a web-grounded answer using a Sonar model. Perplexity searches the live web (or academic papers / SEC filings), synthesizes an answer and returns the text together with its citations, the underlying search results, and optional related follow-up questions.
- Chat Completion (Advanced) Sends a fully custom chat completion request to Perplexity's Sonar API with a complete messages array and every available control: search mode (web/academic/SEC), domain, language, recency and publish/update date filters, image results and image filters, related questions, web search options (context size, search type, user location), structured JSON output via a JSON Schema, and reasoning effort for reasoning models.
Search
1- Search the Web Searches the web with Perplexity's standalone Search API and returns ranked results (title, URL, snippet, publication date and last-updated date) without any LLM answer generation.
Async Sonar
3- Create Async Chat Completion Submits a Sonar chat completion for asynchronous processing and returns immediately with a request ID and status.
- Get Async Chat Completion Retrieves the status and result of an asynchronous Sonar chat completion by its request ID.
- List Async Chat Completions Lists the asynchronous Sonar chat completion requests submitted with your API key, with their IDs, statuses, models and timestamps.
Agent
2- Create Agent Response Generates a response with Perplexity's Agent API, which orchestrates frontier models (Perplexity Sonar, OpenAI GPT, Anthropic Claude, Google Gemini, xAI Grok) with built-in tools: web search, finance search, people search, URL fetching and a code sandbox.
- Get Agent Response Retrieves a previously created (stored or background) Agent API response by its ID. Returns a snapshot with its status (queued, in_progress, completed, failed, incomplete or cancelled), the output items (messages, search results, tool calls) and token usage.
Models
1- List Models Lists the models available for the Perplexity Agent API, including Perplexity's own Sonar models and frontier models from OpenAI, Anthropic, Google and xAI, with their IDs in 'provider/model-name' format and owners.
Embeddings
2- Create Embeddings Generates vector embeddings for up to 512 texts with Perplexity's embedding models. Each text supports up to 32K tokens (120K tokens combined per request).
- Create Contextualized Embeddings Generates contextualized embeddings for chunked documents with Perplexity's contextual embedding models: each chunk's vector is computed with awareness of its surrounding document, improving retrieval quality for RAG pipelines.
Start building with Perplexity
$100 in credits. No card required. Connect in minutes.