Perplexity web answers with domain and recency filters

Perplexity web answers with domain and recency filters

An n8n workflow that sends one question to Perplexity's Agent API with web search switched on. The search is limited to the
domains you list and to the last month. The workflow returns the answer text and the addresses of its sources.

What is inside

• workflow.json: the workflow as n8n exports it, six nodes: four that do the work and two sticky notes. It holds no
  credential.
• README.md: these notes.

What we changed from the original copy

• The credential links are gone. The original listed a Basic Auth and a Header Auth credential, and the request used the
  Header Auth one. Select your own.
• The request went to https://api.perplexity.ai/chat/completions with the sonar model. Perplexity's documentation, read
  on 12 November 2026, says Sonar chat completions support ended on 27 September 2026, and that requests keep working while
  they are reformulated as Agent API requests. It recommends moving to the Agent API, and Mindket's copy calls
  https://api.perplexity.ai/v1/agent.
• effort set to fast takes the place of the sonar model, which is the mapping the documentation suggests. The prompts go
  in instructions and input. The domain list and the month filter sit inside a web_search tool, where the
  documentation places them.
• Some settings have no direct place in the new request and are gone: top_k, return_images, return_related_questions,
  the two penalties and response_format. The documentation also says temperature and top_p may be ignored,
  depending on the model, so they are gone too.
• Clean Output reads the answer from the reply's output list. The answer was in choices[0].message.content before, and
  the sources were in citations. The sources now come from the search results, each as its number and its address.
• The body is built as an object, so a quote in a prompt no longer breaks the JSON. The domain list is trimmed, so a space
  after a comma no longer reaches Perplexity.
• The two notes are rewritten, and the example key in them is a placeholder.
• Nothing else.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• A Perplexity account with credit, and an API key from its settings.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add a Header Auth credential. Set the name to Authorization and the value to Bearer <YOUR_PERPLEXITY_KEY>, with your
   key in place of the placeholder. Select it in Perplexity Request.
3. In Set Params, set system_prompt, user_prompt and domains, a comma separated list.
4. Click Test workflow and open Clean Output.

How a run goes

1. Set Params holds the system prompt, the question and the domain list.
2. Perplexity Request posts them to the Agent API, with web search limited to those domains and to the last month.
3. Clean Output keeps two fields. output is the answer text, and citations lists each source as [number] address.

Change the look

• The question and the instructions: system_prompt and user_prompt in Set Params.
• The domains: domains in Set Params.
• The time window: search_recency_filter in the body of Perplexity Request.
• The depth of the work: effort in the same body.

Good to know

• The domain filter takes up to 20 domains. Use the list as an allowlist
  or as a denylist with a leading - on each entry, and never mix the two.
• The sample list n8n.io, make.com is only an example. To search without a domain filter, delete the search_domain_filter
  line from the body, because an empty list would send one empty entry.
• The recency filter takes hour, day, week, month or year. The file sets month.
• At the fast level the answer carries numbered markers such as [1], and each number matches the same number in
  citations.
• fast suits a quick lookup. For harder questions, the documentation offers low, medium, high and xhigh.
• The original setup note says to buy credit before you start.
• The prompts go to Perplexity.
• The JSON holds no credential.