Chat agent that reports what is trending on YouTube

Chat agent that reports what is trending on YouTube

An n8n chat for people who make videos. You name a niche, and an AI agent searches YouTube for videos from the last two
days, looks each one up for its numbers, and tells you what the new videos have in common. A search tool does the
lookup, and the tool is the lower half of the same workflow.

What is inside

• workflow.json: the workflow as n8n exports it, sixteen nodes. Two are sticky notes, and one was added by Mindket.
  It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The lower half of the file had no start, because it was meant to live in a workflow of its own. Mindket added a
  When Executed by Another Workflow node in front of get_videos1, set to accept all data, so the tool can call this
  same workflow. The note on that half was reworded to say so.
• youtube_search pointed at a workflow in its author's n8n. It now points at none until you pick this one.
• get_videos1 was linked to a YouTube credential in its author's n8n. The link is gone, so select your own.
• Nothing else. Its Code nodes and expressions needed no repair.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenAI API key.
• A Google Cloud project with the YouTube Data API on. It gives you the OAuth2 credential for get_videos1 and the API
  key for find_video_data1.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Save it, so the workflow has an ID of its own.
3. In youtube_search, pick this same workflow from the list. The tool runs the lower half of the file, from
   When Executed by Another Workflow.
4. Add an OpenAI credential and select it in openai_llm.
5. Add a YouTube OAuth2 credential and select it in get_videos1.
6. Give find_video_data1 your YouTube Data API key. On a server of your own, set the environment variable
   GOOGLE_API_KEY for n8n. Where you cannot set one, replace {{ $env["GOOGLE_API_KEY"] }} in the key parameter with
   the key itself, and take it out again before you share the file.
7. Save, open the chat on the canvas and name a niche.

How a request runs

1. chat_message_received passes your message to AI Agent. window_buffer_memory keeps the last five exchanges.
2. The agent checks that you gave a niche. If you did not, it asks and offers a few to pick from.
3. It calls youtube_search up to three times, each time with a different search term. The tool receives one field,
   search_term, which arrives in the lower half as query.search_term.
4. get_videos1 searches YouTube for that term: region US, videos from the last two days, three results, ordered by
   relevance, with moderate safe search.
5. loop_over_items1 takes one result at a time. find_video_data1 fetches its details, with the parts
   contentDetails, snippet and statistics.
6. if_longer_than_3_ passes a video on only when it runs longer than 210 seconds. group_data1 keeps its id, views,
   likes, comments, description, title, channel name, tags and channel id.
7. save_data_to_memory1 strips links and emoji, and adds the video to one text, with a marker between videos.
8. When the loop ends, retrieve_data_from_memory1 and response1 hand that text back to the agent, which writes its
   report.

Change the result

• The report: the system message of AI Agent. As written, it asks for the overall picture, not a review of any one
  video, with the numbers and links.
• Which videos count: get_videos1 holds the two-day window in publishedAfter, the region, the number of results
  (limit) and the order. The 210 seconds sit in if_longer_than_3_.
• How much the chat remembers: Context Window Length in window_buffer_memory.
• The model: openai_llm.

Good to know

• openai_llm names no model in the file, so the node uses its own default, gpt-5-mini (n8n's repository, read on
  12 November 2026). OpenAI's deprecations page, read the same day, lists gpt-5-mini for shutdown on 11 December 2026,
  with gpt-5.6-terra as its replacement. Pick a model in the node before then.
• save_data_to_memory1 builds its text in the workflow's saved data and never clears it. n8n keeps that data between
  runs only on a published workflow, so there a later search can return videos from an earlier one. While you test by
  hand, nothing stays between runs.
• group_data1 joins each video's tags, so a video with no tags stops the run at that node.
• A video of 3 minutes 30 seconds or less is left out, whatever its numbers.
• Each tool call costs YouTube quota: one search, and one lookup for each of up to three videos.
• OpenAI bills the agent's calls, and a niche can take up to three tool calls.
• The chat trigger's public page is off until you turn it on in the trigger. Leave it off unless you want strangers to
  use your keys.
• Your messages and the video data go to OpenAI. The searches go to Google.