Count kci-dev as its own analytics client

[alanpeixinho]

Problem

kci-dev sends User-Agent: kci-dev/<version> and nothing else. get_client_info does not recognize it, so those requests are browser=unknown, os=unknown, and device=desktop. They also increment dashboard_unique_visitors_total, which has no client label, so CLI installs and dashboard users are one number.

In the week of 21–28 Sep 2026, unknown was 65% of OS requests and 50% of browser requests. Much of that is this CLI.

Proposal

In get_client_info, before the other guesses, match a User-Agent that starts with kci-dev followed by / or the end of the string. Set browser=kci-dev, os=kci-dev, device=cli. Do not guess Linux: the User-Agent has no OS.

Add a client label (dashboard or kci-dev) to dashboard_unique_visitors_total and dashboard_unique_visitors_by_endpoint_total.

In monitoring/dashboard.json, filter the unique-visitor stat to client="dashboard" and add a CLI unique-visitor stat for client="kci-dev". The browser, OS, and device panels already group by those labels, so kci-dev and cli show up without new queries. No Prometheus scrape change. The metric labels are the Prometheus change.

Tests

• kci-dev/0.1.11 → browser kci-dev, os kci-dev, device cli, unique visitor only on client="kci-dev"
• a normal browser User-Agent is unchanged and increments only client="dashboard"

[aliceinwire]

I’d suggest adding an explicit client label for dashboard vs kci-dev, while keeping os=unknown for CLI requests. Using os=kci-dev mixes application identity with the OS dimension and would make the Grafana OS panel misleading. device=cli looks appropriate.