Umami weekly traffic summarized by AI and saved to Baserow

Umami weekly traffic summarized by AI and saved to Baserow

An n8n workflow that turns a week of website traffic into a written report. Each week it reads your site's figures from
Umami and asks an AI model on OpenRouter for a summary table. It then reads the top pages of this week and last week and
asks the model to compare them and suggest five changes. Both answers are saved as one row in a Baserow table.

What is inside

• workflow.json: the workflow as n8n exports it, 13 nodes and six sticky notes. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• Get view stats from Umami called event-data/stats, the route for custom events, and the code then looked for page
  views and visitors in its answer. The node now calls stats, which gives page views, visitors, visits, bounces and
  total time.
• The two page nodes asked for type=url and sent a tz setting. Umami's source, read on 12 November 2026, lists path
  among the metric types and has no url, so both nodes now use type=path. They also ask for a short list of pages. Umami's own default is 500, which makes a long question. The time zone setting was not needed and is gone.
• Parse Umami data read each figure as a value and a prev pair. Umami's current answer holds each figure as a
  number, with the 7 days before in a comparison object. The code reads both shapes.
• All three code nodes ran their data through encodeURIComponent, so the model received percent codes. The code nodes
  now write plain text lines, and each AI body is built with JSON.stringify.
• Both AI nodes asked for meta-llama/llama-3.1-70b-instruct:free. OpenRouter's public model list, read on 12 November
  2026, does not list that name. It lists meta-llama/llama-3.1-70b-instruct, so this copy uses it.
• Each page node now has Always Output Data on, so a week with no page views still reaches the compare step.
• A new node, Site to check, holds the Umami address, the website ID and the site name once. The original copy typed
  its author's hosts and website IDs into the nodes, and its prompts named the author's site. They are gone.
• A new node, Row for Baserow, names the four values Date, Summary, Pages and Site, and Save data to Baserow maps them
  to the table's fields by name. The original copy picked fields by number, and those numbers belonged to its author's table.
• The six sticky notes are rewritten and hold no outside links. The original copy linked to its author's pages.
• Every credential reference is removed, so you pick your own.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An Umami site that you can read, its API address, its website ID and a token. On Umami Cloud the address is
  https://api.umami.is/v1 and the token is an API key.
• An OpenRouter API key.
• A Baserow account, on Baserow Cloud or a server of your own, and a table with the fields Date, Summary, Pages and Site.
  Make Summary and Pages long text fields, because the answers are long.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. In Umami, find the website ID in the settings of your site.
3. Get a token. Send a POST request to /api/auth/login on your Umami server, with a JSON body that holds your username
   and password. The answer holds a token.
4. In Site to check, replace YOUR_UMAMI_HOST with the address of your Umami server and keep /api at the end. Replace
   YOUR_WEBSITE_ID with the website ID and YOUR_SITE_NAME with a name for the site.
5. In Get view stats from Umami, add a Header Auth credential with the name Authorization and the value Bearer, a space
   and your token. Select the same credential in Get page data from Umami and Get page view data from Umami.
6. In Send data to A.I., add a second Header Auth credential with the name Authorization and the value Bearer, a space
   and your OpenRouter API key. Select it in Send data to A.I.1 too.
7. In Save data to Baserow, add a Baserow credential with your Baserow address, email and password. Then pick your
   database and your table.
8. Press Test workflow. A new row appears in the table.
9. Turn the workflow on, so that Schedule Trigger runs it every week.

How a run goes

1. Schedule Trigger starts the run by itself. Its rule is weekly on Thursday, and the hour and minute keep n8n's
   defaults, so it is Thursday at 00:00 in the timezone of your n8n. When clicking ‘Test workflow’ starts it by hand.
2. Site to check passes on the three values.
3. Get view stats from Umami reads the figures of the last 7 days.
4. Parse Umami data turns them into text lines, each with the figure for the last 7 days and the 7 days before.
5. Send data to A.I. asks OpenRouter for a summary table and two or three observations.
6. Get page data from Umami reads the top 20 pages of the last 7 days, and Parse Umami data1 turns them into lines.
7. Get page view data from Umami reads the top pages of the 7 days before, and Parse Umami turns them into lines.
8. Send data to A.I.1 sends both lists and asks for a comparison table and five suggestions.
9. Row for Baserow takes the date, the first answer, the second answer and the site name.
10. Save data to Baserow adds them as a new row.

Change the look

• The site: the three values in Site to check.
• The day and the time: the rule in Schedule Trigger.
• The length of the period: the number of days in the addresses of the three Umami nodes.
• The number of pages: limit=20 in the two page nodes.
• The questions: the text in the two AI nodes.
• The model: the model name in the JSON body of both AI nodes. Any model on OpenRouter's list works.
• The columns: the four values in Row for Baserow. Their names must match the table's fields.
• Several sites: make a copy of the workflow for each website.

Good to know

• The model that this copy names, meta-llama/llama-3.1-70b-instruct, costs 0.00000072 US dollars per token for the prompt and the same for the answer on that list, which is about 0.72 US dollars per million tokens. Two calls a week are a small bill. For no bill, name a free model from the same list, such as google/gemma-4-31b-it:free. Free
  models have usage limits, and a free name can leave the list later.
• Umami's documentation says to send the token from the login route as a Bearer header. If a run ends with an
  authorization error, make a new token the same way and update the credential.
• The number next to each page is Umami's count of distinct sessions for that page.
• Your data: page paths, visitor counts and the figures go to OpenRouter and the model's provider. The two notes are
  stored in Baserow.
• The weekly rule only runs while the workflow is on. A run you start by hand works at any time.