A Prometheus metrics exporter for Umami website analytics data. This exporter reads directly from Umami's PostgreSQL database to provide website statistics as Prometheus metrics.

Run the exporter using Docker:

docker run -p 8080:8080 \
  -e DATABASE_URL="postgres://user:password@host:5432/umami?sslmode=require" \
  -e WEBSITE_ID="your-website-uuid" \
  ghcr.io/csmith/umami-exporter:latest

Metrics will be available at http://localhost:8080/metrics.

The exporter is configured using environment variables or command-line flags:

The DATABASE_URL should be a standard PostgreSQL connection string:

postgres://username:password@hostname:port/database?sslmode=require

Find your website ID in the Umami admin interface or by querying the website table in your database.

Counter tracking total page views with the following labels:

• url_path - The page path (e.g., /, /about)
• referrer_domain - Referring domain (empty string for direct visits)
• browser - Browser name (e.g., Chrome, Firefox)
• os - Operating system (e.g., Windows, macOS, Linux)
• device - Device type (e.g., desktop, mobile, tablet)
• country - Two-letter country code (e.g., US, GB, DE)

Histogram showing the distribution of pages viewed per visit, with buckets at 1, 2, 3, 5, 10, 20, 50, and 100 pages.

This exporter can generate metrics with very high cardinality due to the url_path label. If your website has many unique URLs (dynamic paths, query parameters, etc.), this could create thousands or millions of unique metric series.

The exporter queries the entire dataset for the website in question, with no time limit. This may get slower and slower over time.

services:
  umami:
    # ...

  umami-db:
    #...

  umami-exporter:
    image: ghcr.io/csmith/umami-exporter:latest
    environment:
      DATABASE_URL: "postgres://umami:umami@unami-db:5432/umami"
      WEBSITE_ID: "your-website-uuid-here"
      LOG_LEVEL: "info"
    depends_on:
      - unami-db

Pull requests and issues are more than welcome. There are probably better ways to expose the data, more interesting things that could be exported, and so on. I'm open to suggestions!

This project was primarily created with Claude Code, but with a strong guiding hand. It's not "vibe coded", but an LLM was still the primary author of most lines of code. I believe it meets the same sort of standards I'd aim for with hand-crafted code, but some slop may slip through. I understand if you prefer not to use LLM-created software, and welcome human-authored alternatives (I just don't personally have the time/motivation to do so).