EASY Self-Hosted n8n Tutorial | Automate AI Agents, Notion & More!

TL;DR

n8n is free, but it must be hosted on infrastructure such as a VPS; the tutorial recommends a KVM2 plan for handling more simultaneous automations.

Briefing Cornell Notes

Briefing

Self-hosted n8n can turn scattered information—RSS articles, YouTube channels, and Notion task data—into a scheduled daily digest that automatically publishes inside Notion. The core payoff is control: n8n runs on your own server, so automations are transparent, customizable, and able to pull from multiple APIs and tools without relying on a single vendor’s workflow limits.

The walkthrough starts with hosting. n8n is free, but it needs a place to run, so the setup uses a VPS from Hostinger (with a recommended KVM2 plan for more simultaneous automations). After the VPS is provisioned, the process moves into n8n’s “start from scratch” flow, including creating an owner account and selecting the company profile details. Once the instance is ready, the build focuses on a practical AI-agent workflow rather than a toy example.

The first workflow step uses a chat trigger to prove the agent can respond. An AI agent node is then added, wired to the chat input as the prompt. Model flexibility is a major theme: the agent can use chat models from providers like Anthropic, Google Gemini, and OpenAI. The tutorial demonstrates Google Gemini first, including creating a Gemini API key via Google AI Studio and attaching it as an n8n credential. A key limitation appears immediately—without memory, the agent can’t recall prior user details—so a memory option is added to retain the last five interactions. With memory enabled, the agent can answer follow-up questions about earlier messages.

Next comes “tools,” which let the agent fetch external content. An RSS read tool is added for TechCrunch by supplying the RSS feed URL (using a feed URL pattern). Executing the step shows titles, creators, tags, and categories; asking for “latest articles” returns a digest and even direct links. The workflow is duplicated to add another RSS source for YouTube content (Linus Tech Tips), with the feed URL extracted from the channel page source by searching for the channel ID. At this stage, the agent can generate a combined daily digest from multiple sources.

The workflow then connects to Notion to incorporate overdue tasks. A Notion integration is created using an internal integration secret from notion.so, and the agent is granted access to a specific task-tracker database. The Notion node pulls database pages (the tutorial returns the three tasks in the example database). Finally, the output is published back into Notion: a new database is created with a feed view, and the workflow uses a “create a database page” action to insert the daily digest text.

To make it automatic, the chat trigger is replaced with a scheduled trigger so the digest runs every morning. The memory is removed because the workflow no longer depends on a live chat session. The agent’s output is constrained with a prompt (e.g., “2000 character or less,” simple text, no bolding/headings), and the persona is set via a system message (“snappy and fun assistant writer”). One practical caveat is noted: Notion links may require manual insertion via rich text blocks. After saving and activating the workflow, the digest starts running. The tutorial closes with tips for fixing formatting using Claude or ChatGPT and mentions a personal use case—generating scopes of work from discovery-call transcripts—plus a reminder to save workflows to avoid losing changes.

Cornell Notes

The tutorial builds a self-hosted n8n automation that generates a daily AI digest and publishes it to Notion. It starts by hosting n8n on a VPS, then creates an AI agent workflow that can use different model providers (demonstrated with Google Gemini) and optional chat memory. “Tools” are added through RSS readers to pull the latest TechCrunch and YouTube (Linus Tech Tips) items, and a Notion integration pulls overdue tasks from a task-tracker database. The workflow is converted from a chat-based trigger to a scheduled trigger so it runs every morning, then writes the formatted digest into a Notion database page. This matters because it shows how to combine APIs, AI, and scheduling into a controllable, repeatable system.

Why does the workflow need “memory,” and what changes when the trigger switches from chat to scheduling?

Memory is used when the agent is driven by a chat trigger, because follow-up questions depend on earlier user messages. In the demo, the agent can answer “my favorite food is mac and cheese,” but without memory it can’t recall that detail when asked later. Adding memory (set to remember five past interactions) enables that recall. Once the workflow is changed to a scheduled trigger, there’s no ongoing chat session to retain context, so memory is removed to avoid unnecessary state and to keep the run deterministic.

How does n8n let the AI agent access external information beyond its prompt?

n8n adds “tools” that the AI agent can call. The tutorial uses an RSS read tool to fetch article/video metadata from URLs. For TechCrunch, it supplies the RSS feed URL and the tool returns titles, creators, tags, and categories; the agent then summarizes “latest articles” and can output links. A second RSS read tool is duplicated for YouTube by extracting the channel feed URL from the channel page source (searching for channel_id). This turns the agent into a content aggregator rather than a text-only chatbot.

What steps are required to connect Notion data (overdue tasks) into the workflow?

The workflow creates a Notion integration using notion.so integration settings, then copies the integration secret into n8n as a credential. It also configures access so the integration can read the specific task-tracker database. In the Notion node, it selects the database and uses actions like “get many” / “database pages” to retrieve the task items (the example returns three tasks). Those retrieved pages become part of the agent’s input for generating the daily digest.

How does the digest get published back into Notion automatically?

The tutorial creates a separate Notion database with a feed view (e.g., “my daily news from N8N”). It then grants the integration access to that new database. In n8n, the workflow uses a Notion action like “create a database page,” targeting the correct database. The agent’s generated digest text is inserted into the page content (via blocks such as rich text), and the workflow executes to verify the page is populated.

What practical limitation appears when adding links inside Notion, and how is it handled?

Notion connections require links to be added manually in the rich text block. After the agent generates the digest, the workflow may need an explicit rich-text step where the user provides link text and pastes the database URL (e.g., linking “access your tasks” to the task tracker). The tutorial demonstrates copying the task database link from Notion and inserting it into the rich text block, then executing to confirm the clickable link.

How does the tutorial manage formatting and length of the AI output for Notion?

Formatting is controlled through the prompt passed to the AI agent in the scheduled workflow. The tutorial instructs the agent to produce a digest “2000 character or less” and specifies output constraints like “simple text, no bolding, no headings.” If output still doesn’t match the desired structure, the tutorial suggests using Claude or ChatGPT to generate or repair code/formatting, including a workflow where screenshots of the n8n setup are used to ask for fixes.

Review Questions

When would chat memory be necessary in an n8n AI agent workflow, and what symptom would appear without it?
Describe the chain of components needed to go from external content (RSS) to a Notion page (integration → database read → AI digest → database page write).
What changes when replacing a chat trigger with a scheduled trigger, and why does that affect memory usage?

Key Points

1
n8n is free, but it must be hosted on infrastructure such as a VPS; the tutorial recommends a KVM2 plan for handling more simultaneous automations.
2
A chat-triggered AI agent can be tested interactively, then upgraded into a scheduled workflow for unattended daily runs.
3
Model choice is flexible in n8n; the tutorial demonstrates Google Gemini via an API key created in Google AI Studio and notes OpenAI as an alternative with usage-based costs.
4
RSS read tools let the agent fetch and summarize the latest items from websites and YouTube channels by using feed URLs.
5
Notion integration requires creating an integration secret, granting access to the relevant databases, and using nodes to read overdue tasks and write digest pages.
6
Publishing into Notion is done by creating a database page in a target database, with the agent’s output inserted into blocks such as rich text.
7
Notion links may require manual insertion in rich text blocks, even when the rest of the digest is automated.

Highlights

Self-hosted n8n can automatically generate a morning digest by combining RSS content, AI summarization, and Notion task data.

Adding memory (five past interactions) enables follow-up questions in chat; removing memory is appropriate once the workflow becomes scheduled.

RSS feeds power the “latest articles/videos” tool, including YouTube channel feeds extracted from page source.

The Notion loop is end-to-end: read overdue tasks from one database, then create a new Notion page in another database with the formatted digest.

Notion link formatting can require manual rich-text link insertion even when the digest text is AI-generated.

Topics

Self-Hosting n8n
AI Agents
RSS Feeds
Notion Integrations
Scheduled Automations