Perplexity research turned into an HTML web page

Perplexity research turned into an HTML web page

An n8n workflow you call with a web address. You pass a topic in the address, an AI agent has Perplexity research it, and the workflow answers with a finished HTML article page.

What is inside

• workflow.json: the workflow as n8n exports it, forty-seven nodes. Thirty-eight do the work and nine are sticky notes that explain the steps. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The Perplexity node called Perplexity's Sonar chat completions with the model llama-3.1-sonar-small-128k-online. Perplexity's docs, read on 12 November 2026, say Sonar Chat Completions support ended on September 27, 2026 and name POST https://api.perplexity.ai/v1/agent as the replacement. Mindket's copy calls that address with effort set to fast, a web_search tool limited to the last month and 4000 output tokens. Success Response now reads the answer from the output list.
• The original limited the search to the domain perplexity.ai itself. Mindket's copy drops that limit, so the search covers the web. To limit it again, add search_domain_filter to the tool's filters.
• The original's sampling options and its flags for images and related questions are not sent, because the Agent API has no direct field for most of them.
• The original pasted the topic into a JSON string, so a quote in the topic broke the call. The copy builds the body as an object.
• The tool node Call Perplexity Researcher pointed at the original author's workflow by its id. That field is empty now. Save this workflow once and pick it there, because the lower row is the researcher the tool calls.
• The Telegram chat id was a sample number in a Number field. It is now the placeholder <YOUR_TELEGRAM_CHAT_ID>.
• The original copy had stray text after its JSON, which Mindket's copy drops. Several nodes were linked to credentials in the author's n8n. The links are gone, so select your own.
• Nothing else. The prompts and the sticky notes are the file's own.

Before you start

• An n8n instance with a public address, on n8n Cloud or your own server, because the workflow starts from a webhook.
• An OpenAI API key.
• A Perplexity API key.
• A Telegram bot and the id of the chat it writes to. This part is optional.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json. Save it once.
2. Open Call Perplexity Researcher and pick this same workflow in its Workflow field. It runs the lower row, which starts at Execute Workflow Trigger.
3. Add an OpenAI credential and select it in the five model nodes, gpt-4o-mini, gpt-4o-mini1, gpt-4o-mini2, gpt-4o-mini3 and gpt-4o-mini5.
4. Choose a model in each of those nodes. The file names gpt-4o-mini-2024-07-18 in the first and sets none in the other four. OpenAI's deprecations page, read on 12 November 2026, says gpt-4o-mini-2024-07-18 shuts down on 14 January 2027 with gpt-6-luna as the replacement. It also says gpt-5-mini, which current n8n uses when none is set, shuts down on 11 December 2026 with gpt-5.6-terra as the replacement. The model behind Perplexity Topic Agent calls a tool, so pick one with function calling, such as gpt-5.6-terra. If a model refuses the temperature option, clear it in that node's options.
5. Add a Header Auth credential for Perplexity. Name it Authorization and give it the value Bearer <YOUR_PERPLEXITY_KEY> with your own key. Select it in Perplexity.
6. For Telegram, add a Telegram credential and select it in Telegram and Telegram2. Replace <YOUR_TELEGRAM_CHAT_ID> in Chat Id and Chat Id1 with your chat's id. Without Telegram, disable those two nodes. A disabled node passes its input on.
7. Open the Webhook node and copy its test address. Add ?topic= and a topic to it, listen for a test event and open the address in a browser.

How a run goes

1. Webhook receives the request, and Get Topic takes query.topic. If Topic Exists stops an empty topic.
2. Improve Users Topic (Basic LLM Chain 1.4) rewrites the topic as a prompt of at most two sentences for Perplexity.
3. Perplexity Topic Agent (Agent 1.6) is told to use perplexity_research_tool, which is Call Perplexity Researcher.
4. The tool starts the lower row. Execute Workflow Trigger passes the topic to Chat Id1 and Telegram2, which posts it to Telegram. Then If checks that a topic exists. After it, Prompts sets the system and user texts, and Perplexity (HTTP Request 4.2) asks the Agent API. Last, Success Response returns the text.
5. Back in the upper row, Chat Id and Telegram send the first 300 characters of the research to Telegram. If2 and Extract JSON then turn the research into an article object. It has a category, a title, details, a main text, sections with quotes and hashtags.
6. Article and If Article pass it on. Create HTML Article converts it to a title and one line of HTML, and If HTML and Contents pick the two fields.
7. Basic LLM Chain turns that into a full page styled with Tailwind CSS, and Respond to Webhook sends it back as text.

Change the result

• The research prompt: Prompts and Improve Users Topic.
• Search options: the body of Perplexity, including effort, search_recency_filter and max_output_tokens.
• The article's shape: the example schema in Structured Output Parser1.
• The page's look: the prompt in Basic LLM Chain.
• The webhook's path: pblog in Webhook.

Good to know

• An empty topic ends in a Do Nothing node, so the caller gets no page back. Those branches never reach Respond to Webhook.
• If2 and If Article compare text written without the leading =, so they always pass.
• The caller waits for the whole chain, which makes several model calls in a row.
• The page it returns loads Tailwind CSS 2.2.19 from jsDelivr.
• Perplexity and OpenAI bill their calls to your keys.
• The topic goes to OpenAI and Perplexity. Telegram gets the topic and the first 300 characters of the research.
• Anyone who knows the webhook's address can start a run on your keys once the workflow is published.
• The file holds no credential and no key.