Features
What Lynx keeps an eye on for you.
Ten parts, one interface: the agent on your servers, checks from outside, alerts, forecasts, status pages, the weekly report, the AI help, security findings, web traffic analysis and separate areas for your clients.
Every feature in detail
01
Servers from the inside
A small agent measures every minute how your servers are doing — and spots bottlenecks before they turn into outages.
The agent is a single program without dependencies. It runs under its own unprivileged account, opens no port and makes one outgoing connection over HTTPS. If the connection drops, it buffers up to 24 hours of measurements and delivers them later.
Every minute it records CPU per core, load, memory and swap, disk and network throughput, the fill level and inodes of every file system, and the state of MySQL, PHP-FPM and Apache. The ten processes using the most CPU and the ten using the most memory appear by name.
On top of the raw numbers, Lynx runs checks that know what a web server looks like when it is about to struggle: Apache workers running out, a PHP-FPM pool at its limit, MySQL refusing connections, a backup that is older than it should be. The agent currently supports Ubuntu 20.04 to 26.04.
Metrics
CPU, load, memory, disks, network — every minute
Services
MySQL, PHP-FPM and Apache
Top processes
by name only, never with a command line
Bottlenecks
Apache workers, PHP-FPM pools, MySQL connections, backup age
02
Websites from the outside
Four check locations in Nuremberg, Helsinki, Ashburn (USA) and Singapore fetch your pages. An alert is raised only when the locations agree.
A monitor is a question asked from the outside, once a minute by default: does the page answer over HTTPS, with the expected status, and does it contain the text it should? You can send your own headers, log in with basic or bearer authentication and require that a certain text is absent.
Certificates and domains expire quietly. Lynx warns 14 days and again 3 days before a certificate runs out, and 30 and 7 days before a domain does — for every top-level domain whose registry publishes the expiry date. Some registries, the one for .de among them, do not.
Not everything is a website. Ping, TCP and DNS monitors cover mail servers, databases and name resolution. A heartbeat monitor turns the question around: your cron job or backup calls an address when it has finished, and Lynx alerts you when the call does not arrive in time.
HTTPS and keyword
is the right content on the page?
Certificate and domain
a warning well before expiry
Ping, TCP, DNS
for everything that is not a website
Heartbeat
cron jobs and backups report in themselves
03
Alerts
Notifications where your team works — debounced, with quiet hours, escalation and a clear all-clear.
A check has to fail twice in a row before it becomes an incident, and a monitor from the outside only counts as failed when every location that is reporting sees it. A check that flaps — more than three changes within an hour — produces one notification, not a dozen.
You decide who hears about what, and when. Quiet hours follow your time zone and can let serious incidents through. If nobody acknowledges an incident, Lynx notifies the next step after 15 and 30 minutes. A maintenance window holds notifications back for the whole account, one server or one monitor while measuring continues.
When a problem is over, the people who were alerted get the all-clear, with the duration. And Lynx tells you when it can no longer see: a check that cannot measure is reported as blind, and an agent that has been silent for ten minutes becomes an incident of its own.
Channels
email, Telegram, push, Slack, Microsoft Teams, webhook
Quiet hours and escalation
peace at night, and nobody who fails to react is forgotten
All-clear
with the duration, to everyone who was alerted
Maintenance windows
notifications held back, measuring continues
04
Forecasts
“Disk full in ≈ 9 days” — Lynx reads the trend, so you can act during working hours.
Every hour Lynx fits a trend line through the last seven days of each file system: how fast is it filling up, and when will it be full? The same is done for inodes, and optionally for memory. The method is robust against outliers, so a single large upload does not trigger a warning.
A forecast only becomes a warning when the trend is clear and holds for two runs in a row: less than seven days until full is a warning, less than two days is serious. A disk that is merely full of static data stays quiet — there is no trend to report.
Disk
per file system, days until full
Inodes
the limit that many small files hit first
Memory
steady growth, switched on if you want it
Two thresholds
warning under 7 days, serious under 2
05
Status pages
Show your clients openly that everything is running — or that you are already working on it.
A status page shows the monitors you choose, their current state and — if you want — their availability over the last 7, 30 or 90 days. It shows no server details, no software versions, no ports and no causes: your clients see whether it works, not how it is built.
Put the page on your own domain with two DNS records. Lynx checks them, obtains the certificate from Let’s Encrypt and renews it. The page carries noindex, so it is found through your link and not through a search engine.
Announce maintenance in advance and mute the alerts for its duration. Visitors can subscribe by email with a confirmation link and unsubscribe with one click; Lynx stores the address and the time of the confirmation, nothing else.
Your own domain
status.your-company.com with an automatic certificate
Announce maintenance
in advance, with alerts muted meanwhile
Email subscription
visitors hear about incidents automatically
History
availability over the last 90 days

06
Weekly report
Every Monday at 07:00: the past week in one email — short, with numbers.
The report opens with what needs attention right now, then looks back: availability of your monitors, the number and duration of incidents, false alarms, and how each compares with the week before. A number that does not exist is left out rather than shown as zero.
It closes with what is coming up in the next 30 days: disks that are filling, certificates and domains that expire, open security findings. It is switched on for the account owner; anyone else can subscribe, and everyone can unsubscribe with one click.
Availability
of all monitors, compared with the previous week
Incidents
how many, how long, the most serious first
Coming up
disks, certificates and domains within 30 days
Unsubscribe
one click, no login
07
AI help and MCP
With your own Anthropic key, Claude analyses incidents and suggests the first step. Switched off until you agree.
When an incident opens, Claude can read what Lynx measured and write a short assessment: the likely cause and where to start. It never runs anything on your servers. A detailed analysis is available at the push of a button on the incident.
The AI help is switched off until the account owner switches it on and confirms what is sent. It uses your own Anthropic key, within a monthly limit you set. File contents, command lines, user names and credentials are never sent; names and addresses can be replaced before sending.
Through MCP you can ask Lynx from claude.ai, Claude Desktop or Claude Code: which servers have a problem, what happened last night? The access is read-only, has to be enabled by the owner and is authorised per user.
The AI incident analysis in detailMore about security and privacyFrom the blog: Monitoring
Summary in the alert
likely cause and first step in a few sentences
Detailed analysis
at the push of a button on the incident
Cost limit
you decide how much it may cost per month
MCP access
ask Lynx directly from claude.ai, Claude Desktop or Claude Code
08
Security findings
Once a night Lynx compares what is installed on your servers with what is known to be vulnerable.
The agent reports the installed packages; the central matches them against public vulnerability databases and the security notices of the distribution. A finding is rated critical only when the affected service is measurably reachable from the outside — so the list starts with what matters.
Software that has reached its end of life is listed too, and a daily check looks for files and folders in the web root that anyone can write. Pending updates, pending security updates and a required restart appear in the facts of each server. Findings never appear on a public status page.
Known vulnerabilities
in installed packages, checked every night
End of life
software that no longer receives fixes
Web root permissions
files and folders that anyone can write
Pending updates
security updates and required restarts per server
09
Web traffic
Who visits your pages – people, search engines, AI crawlers or attackers – counted on your own server and switched on per server, by you.
The agent reads the access log of Apache or nginx and counts per minute and page: requests, visits, bytes sent and status codes. Only these totals go to Lynx – no IP addresses, no user agents, no single requests.
It is off until the account owner switches it on for a server and confirms that their privacy policy covers it. You can switch it off at any time, and delete the counted data straight away if you wish.
People and bots
five classes, by name and behaviour
Pages and errors
top pages, error pages with referrer, downloads
Totals only
no IP address leaves the server, not even as a hash
Switched on by you
per server, by the owner, with a clear explanation
010
Agencies and sub-tenants
One account for all your clients — each in a separate area, with their own access.
A sub-tenant is a closed area for one client: their servers, monitors, alert channels and status page. You see all of your clients; a client sees only their own area. The separation is enforced in the database, not only in the interface.
Four roles decide who may do what: the owner manages users and installs agents, technicians acknowledge incidents and change alert rules, readers see everything and change nothing, and end customers see their own servers and incidents.
Sub-tenants
one closed area per client
Roles
owner, technician, reader, end customer
Status page per client
on the client’s own domain
One invoice
everything counts towards the agency’s plan
Ready for the first heartbeat?
One command per server, your websites as monitors, notifications where your team already works.