n8n AI Agent That Reads the News and Emails You a Summary Every Day
Learn how to build an n8n AI agent that pulls news from RSS feeds, filters it by topic, and sends you a clean email summary every morning automatically.
Every morning at 7 AM, my inbox receives a neatly formatted HTML email with the day’s top stories in AI, software, hardware, and cybersecurity — curated by an AI agent, not by me. I built this in n8n in under an hour, and in this article I’ll walk you through exactly how it works. Get the workflow here
What the Workflow Does
The agent wakes up on a schedule, pulls articles from two RSS feeds (TechCrunch and Wired), filters out anything that isn’t tech-relevant, deduplicates stories that appear across both feeds, ranks them by importance, and sends a clean HTML digest to your inbox.
The whole thing runs on five nodes.
The Five Nodes
1. Schedule Trigger Fires the workflow every morning at 7:00 AM. You set this once and forget it.
2. AI Agent This is the brain. It receives a simple prompt — “fetch today’s news, summarize it, send the digest” — and it decides how to do that using the tools connected to it. The system prompt gives it strict instructions: call every RSS tool, filter by topic, deduplicate, rank, then send the email. It is not allowed to stop until the email is sent.
3. RSS Feed Tools (×2) Two RSS reader nodes are connected to the agent as tools — one pointed at TechCrunch, one at Wired. The agent calls both, collects everything returned, and processes the combined list.
4. Email Tool (Gmail) Once the digest is assembled, the agent calls this Gmail tool to send the finished email. The subject line and body are generated by the AI from the digest it just wrote. This step is mandatory — the system prompt explicitly tells the agent it is not done until this tool is called.
Choosing the Right LLM
The workflow was originally set up with the NVIDIA Nemotron Chat Model as the language model. In practice, if you run into version compatibility issues with that node in your n8n instance, swapping in GPT-4o mini works cleanly — it handles the multi-tool orchestration well and keeps costs very low given the task is purely text-based.
The workflow file includes an OpenAI Chat Model node as an alternative. You just connect whichever one you want to the AI Agent’s language model slot.
Why the System Prompt Is Doing Most of the Work
The most important part of this workflow isn’t the nodes — it’s the instructions inside the AI Agent’s system prompt. A few things worth highlighting:
- “Call ALL RSS feed tools — do not skip any.” Without this, agents sometimes take a shortcut and only call the first tool they find.
- “You are not done until you have called the Email_Tool.” Agents can terminate after producing the summary if you don’t make the sending step explicit and mandatory.
- The output format is defined as HTML, not markdown. Gmail renders HTML correctly. Markdown does not render in most email clients.
These are hard-learned lessons. Getting the system prompt right is the real skill in agentic workflows.
Extending This Workflow
A few directions worth exploring from here:
- Add more RSS feeds. The Hacker News feed, MIT Technology Review, and The Verge all have public RSS endpoints. Each is just another RSS tool node connected to the agent.
- Add a Slack or WhatsApp tool. Send the digest to a channel instead of (or in addition to) email.
- Filter by keyword. Add an instruction to the system prompt to only include stories containing specific keywords relevant to your work — “LLM”, “open source”, “Africa”, whatever fits.
- Store digests. Connect an Airtable or Google Sheets tool so every digest is logged with its date for future reference.
