kittiewake
Get access
ENRU

How it works

A small agent, checks from many cities, alerts you can trust

Kittiewake looks at your servers from two sides: from the inside, with a small agent that reports how each server is doing, and from the outside, with probes in 24 cities that use your site the way your customers do.

The agent

The agent is one small program for Linux with systemd. It runs as a service under its own unprivileged user, reads the numbers Linux already keeps, and sends them to us every 10 seconds over an encrypted connection.

It uses about 15 MB of memory and well under one per cent of a CPU, so it is happy on the smallest VPS.

  • Runs onLinux, systemd
  • Machinesx86-64, ARM64
  • Memoryabout 15 MB
  • Reportsevery 10 s
  • ConnectionTLS 1.3
  • Accessone token per server

Installing

Your dashboard shows a one-line install command with your own token. Run it on the server and the server appears on your dashboard within a minute. Each server gets its own token, which you can revoke at any time.

What it sends, and what it never sends

It sends

  • CPU load and use per core
  • Memory and swap
  • Disk space and disk activity per mount
  • Network traffic per interface: amounts, not contents
  • Whether the services you choose are running, for example nginx or PostgreSQL
  • Uptime, kernel version and the name of the distribution

It never sends

  • The contents of files, databases or logs
  • Anything your customers send you
  • Passwords, keys or environment variables
  • Users or processes beyond the services you named
  • Your network traffic itself

The agent only reports. It cannot run commands sent from us, and nothing on your server changes because of Kittiewake.

Checks from 24 cities

Outside checks need no agent. Give us an address or a port, and probes in 24 cities take turns checking it every 30 seconds: an HTTPS request with its status and the time to the first byte, the TLS certificate and its expiry date, and a TCP connection to each port you list.

One failed check is not an outage. When a city cannot reach you, three other cities try again within ten seconds. Only when they agree do we call it down. When a single city keeps failing, we tell you the trouble is on the way to you from there, not on your server.

What a confirmed outage looks like

  1. Frankfurt — no answer in 10 s
  2. Warsaw — no answer in 10 s
  3. Stockholm — no answer in 10 s
  4. Madrid — no answer in 10 s
  5. Down, confirmed by 4 cities. Alert sent
  6. Frankfurt — 200 OK in 190 ms
  7. Back up after 5 min 31 s. Recovery sent

Probe cities

Europe

  • Amsterdam
  • Frankfurt
  • London
  • Paris
  • Stockholm
  • Helsinki
  • Tallinn
  • Warsaw
  • Bucharest
  • Belgrade
  • Milan
  • Madrid

Türkiye, Caucasus and Central Asia

  • Istanbul
  • Tbilisi
  • Yerevan
  • Almaty

Middle East and Asia

  • Dubai
  • Singapore
  • Tokyo
  • Seoul

Americas

  • New York
  • Toronto
  • Chicago
  • São Paulo

Alerts that respect sleep

Alerts go to email, Telegram or both. Each one says what is wrong, since when, what we checked and what we think it means, in a sentence rather than a code.

  • Urgent — the site is down from several cities, a server stopped reporting, a disk is full. These reach you at any hour, once, and again only if nothing changes for an hour.
  • Can wait — a disk filling up, a certificate that expires in 14 days, memory slowly growing. During your quiet hours these wait until morning.
  • Resolved — when things recover you get one short message, so you can go back to sleep.
DegradedTelegram · 08:00

Disk on db-1 is 87% full

At this rate it fills up in about 9 days. The largest growth is in the database's own files, not in logs.

Noticed at 02:40 and held through your quiet hours (23:00–08:00).

Your metrics stay yours

Metrics are stored in the EU, encrypted at rest, and kept for as long as your plan says: 30 days on the beta, 90 days on Small, a year on Team. We do not sell them, share them or use them to train anything.

The three of us can look at your dashboard only to help when you ask. When you delete a server, its history is gone within seven days, and from backups within 30 days.

Removing the agent

  1. Open the server's page in your dashboard and copy the one-line removal command.
  2. Run it on the server. It stops the service and deletes the program, its configuration and its user. Nothing of Kittiewake is left behind.
  3. Press Delete server. The token stops working at once, and the history is erased.

You do not need to ask us or give a reason.

Try it on your own servers

Kittiewake is in open beta, by invitation, and free while the beta lasts.

Get accesshello@kittiewake.com