Lynx Heartbeat is a monitoring service for Linux servers and websites. An agent measures each server from the inside every minute; four check locations in Nuremberg, Helsinki, Ashburn (USA) and Singapore test pages from the outside. Both appear in one interface, with debounced alerts, forecasts and status pages. The service starts at €9 a month (net, plus VAT) and is hosted in Germany. This article is based on the public product pages, as of 22 November 2026. It is not a hands-on test of an account.
Disclosure: Lynx Heartbeat is a product of CATAMA Software. The provider is Agentur 404Media, owned by Stefan Pingel and based in Potsdam, Germany – the same agency that publishes code404zero. The facts in this article come from the service’s public pages, not from an independent review by a third-party newsroom.
What Lynx Heartbeat is
The about page describes the starting point like this: anyone who looks after servers and websites knows the night an alert arrives, and the question of whether the problem is real and where to start. The load often sits on several tools, one for metrics, one for websites, one for certificates. Lynx Heartbeat is meant to show both in one interface: the state of the servers from the inside and the reachability of the sites from the outside.
The offer is for businesses, not for private individuals. The site names online shops, companies with their own Linux servers, hosting providers and system houses, and agencies that look after servers and websites for many clients. Small teams without an operations department are also in scope: the smallest plan includes 3 servers and 10 monitors.
The view from the inside: the agent on the server
According to the feature description, the agent is a single program with no dependencies. It runs under its own user without special privileges, opens no port and establishes an outgoing connection over HTTPS on port 443. If the connection drops, it buffers measurements for up to 24 hours and delivers them later. It will not start as root.
Every minute it records, among other things:
- CPU per core, including I/O wait, and load, measured per core
- memory, swap and out-of-memory kills
- fill level, inodes and throughput per file system or device, plus network throughput
- the state of services such as Apache, Nginx, PHP-FPM, MySQL, Redis, cron and SSH
- for MySQL, connections, refused and aborted connections, and slow queries
- for PHP-FPM, active processes, the queue and max_children per pool
- for Apache, busy and idle workers and requests per second
- the error rate in logs and newly appearing fatal errors
- the ten processes using the most CPU and the ten using the most memory, by name only, never with the command line
On top of the raw numbers, the agent watches for typical bottlenecks just before a web server buckles: Apache running out of workers, a PHP-FPM pool hitting its limit, MySQL refusing connections, or a backup older than it should be. PHP-FPM in particular is where load and security updates often meet on web servers; which flaws the PHP project recently closed in PHP-FPM is covered in our article on PHP 8.6 and the security updates.
The agent currently supports Ubuntu 20.04, 22.04, 24.04 and 26.04 with systemd, on amd64 and arm64. The installer deliberately rejects other distributions because they have not been released yet. Outside checks do not depend on that: they work for any website or service reachable from the internet, whatever it runs on.
The view from the outside: four check locations
Four locations fetch the public addresses of your monitors: Nuremberg, Helsinki, Ashburn (USA) and Singapore. A website counts as down only when every check location that is currently reporting sees the failure. A network hiccup at a single location does not raise an alert.
By default, a monitor runs once a minute. Options include:
- HTTPS: does the page answer, and with which status?
- keyword: is the expected text on the page, or a text that must not be there?
- certificate: a warning 14 days and again 3 days before it expires
- domain expiry: a warning 30 and 7 days ahead, where the registry publishes the date. Some do not, including the registry for .de.
- ping, TCP and DNS, for example for mail servers, databases and name resolution
- heartbeat: a cron job or backup calls an address when it is done; a missing call is the alert
Custom headers and basic or bearer authentication are supported. According to the provider, the checkers only fetch the public addresses of the monitors and send the results to the central system.
Alerts worth taking seriously
A single failed measurement is not an incident. A check has to fail twice in a row before anyone is notified, and has to be green twice before the all-clear. A check that flaps, meaning more than three changes in an hour, produces one notification rather than a dozen. When the incident is over, everyone who was alerted receives the all-clear, including the duration.
Quiet hours follow your time zone. By default Lynx holds notifications back at night from 22:00 to 06:00 and still lets critical incidents through. If nobody acknowledges an incident, the service notifies the next stage after 15 and after 30 minutes by default. Acknowledging stops the clock. A maintenance window can cover the whole account, one server or one monitor: Lynx keeps measuring and holds the notifications back.
Lynx also tells you when it can no longer see. A check that can no longer measure is reported as blind. An agent that stays silent for ten minutes is an incident of its own.
Without a rule, Lynx notifies nobody and says so. Channels are email, Telegram and push on every plan; Slack, Microsoft Teams and webhook from the Team plan upward. A test message shows whether the chosen path arrives.
Forecasts, status pages and the weekly report
Every hour Lynx fits a trend line through the last seven days of each file system: how fast is it filling, and when will it be full? The same happens for inodes and, if you switch it on, for memory. A single large upload is not supposed to trigger a warning. A forecast becomes a warning only when the trend is clear and holds for two runs in a row: less than seven days to full is a warning, less than two days is serious. The disk forecast needs 24 hours of measurements. A comparable early warning for disk space now also exists inside MySQL itself; what the Health Monitor in MySQL 26.10 does there is covered in our overview of MySQL 26.10.
A status page shows the monitors you select, their state and, if you want, their availability over the last 7, 30 or 90 days. It does not show server internals, software versions, ports or causes. With two DNS records the page can live on your own domain; Lynx obtains the certificate from Let’s Encrypt and renews it. The page is marked noindex. Maintenance can be announced in advance, with alerts muted for its duration. Visitors subscribe to notices by email with a confirmation link. A status page on your own domain is included from the Team plan; the Start plan includes one status page without a custom domain.
Every Monday at 07:00 a weekly report arrives by email: what needs attention now, then availability, the number and duration of incidents and false alarms compared with the previous week, and finally what is due in the next 30 days, such as disks filling up, certificates expiring and open security findings. The report is on for the account owner; everyone else can subscribe.
Security findings and data in Germany
Once a night Lynx compares the installed packages with public vulnerability databases and the distribution’s security notices. A finding is classed as critical only if the affected service is demonstrably reachable from the outside. The list also covers software that has reached end of support, world-writable files and directories in the web root, outstanding security updates and a required reboot. These findings never appear on a public status page. On typical web servers, applying those updates is still the decisive step, for example after a WordPress security release.
According to the provider, the central system runs at Hetzner in Nuremberg, and the backup is stored encrypted off-site in Falkenstein; restoring from it has been tested. Data leaves Germany only when you set that up, for example as a notification to a channel you choose, or through the optional AI help with your own key. The agent sends measurements and check results, never file contents, command lines, environment variables or IP addresses of site visitors. Processes appear by name only.
Sign-in can be protected with an app code and, on the Team and Agency plans, made mandatory for the whole team. On the Agency plan, agencies manage clients as sub-accounts: their own servers, monitors, alert channels and status page per client, with separate access for end customers. The separation is enforced in the database, not only in the interface.
Optionally, and only when the owner enables it per server, Lynx can evaluate web traffic. The agent reads the Apache or Nginx access log and sends only totals: requests, visits, bytes and status codes, no IP addresses, not even as a hash.
How to set it up
You sign up yourself: enter your company, your name and your email address, confirm the address with a code and set a password of at least twelve characters. The code is valid for 24 hours. The account is then ready for 7 days with every feature and without payment details. During the trial, according to the setup guide, you can monitor up to 10 servers and 50 monitors. Each email address gets one trial every 180 days.
You install the agent with one command per server. A --precheck run first tells you whether the server is supported; it changes nothing and needs neither root nor a token. The installer itself needs root because it creates the catama-lynx user and the service. The install token is single-use and valid for 24 hours. The installer waits up to 45 seconds for the first reply; on a test server of the provider, setup finished in under a minute. Agent updates are signed (Ed25519), roll out in stages and can be turned off at install time. Uninstallation is also a single command.
Without a notification rule, nothing is sent. The welcome page also leads to the monitors: if the server has a public name with a valid certificate, Lynx creates an HTTPS monitor from the outside. You add further sites yourself, including heartbeat addresses for cron jobs.
What it costs
There are three plans, all prices net plus VAT. Paying yearly is billed as ten months for twelve, so two months are free. The same discount applies to each extra server.
| Plan | Price per month | Price per year | Servers / monitors | Status pages | Each extra server per month |
|---|---|---|---|---|---|
| Start | €9 | €90 | 3 / 10 | 1 | €3.00 |
| Team | €29 | €290 | 15 / 50 | 3, with a custom domain | €2.00 |
| Agency | €79 | €790 | 50 / 200 | one per client | €1.50 |
Every plan measures the same core set of features, including the agent, the four check locations, 90 days of per-minute check results and then hourly values for two years, forecasts, and alerts by email, Telegram and push. Slack, Microsoft Teams, webhook, status pages on a custom domain, the AI summary in the alert and a mandatory second factor for the team start with Team. Sub-accounts, end-customer access and MCP access for Claude are limited to the Agency plan.
You can add further servers at any time; the price is shown beforehand. For monitors and status pages, the included amount is the limit; for more, you move to the larger plan. Only enabled monitors and published status pages count. At the end of the trial you book a plan or delete the account. Without a plan, monitoring pauses, and the account is deleted with all data 14 days after the trial ends; Lynx sends a reminder by email first.
Optional AI help and MCP
The AI help is off until the account owner switches it on. It uses your own Anthropic key and a monthly limit you set; billing is directly between you and Anthropic. When an incident opens, Claude can write a short assessment of the likely cause. According to the provider, the AI never executes anything on the monitored servers. File contents, command lines, usernames and credentials are not sent; names and addresses can be replaced before sending. The short assessment in the alert is included from the Team plan.
Through MCP you can query Lynx from claude.ai, Claude Desktop or Claude Code, for example which servers have a problem. Access is read-only, must be enabled by the owner and granted per user, and is reserved for the Agency plan.
If you want to look at the service yourself, the current features, the detailed setup and the prices are on lynx-heartbeat.com, including a 7-day trial with no payment details.




