Monitoring for Kubernetes, Docker and Linux servers, without running it yourself.
Install one agent. WatchLog Cloud runs the backend, keeps your data and tells you in plain words why something broke: every node and pod, every systemd service and process (PHP, Node, Go, Java, Python, anything), every container, and the traffic between them.
Prefer your own servers? Self-host the same product for free (AGPL-3.0, no limits).
From signup to your cluster or server
- Sign up with GitHub, Google or email
About a minute. No card, no company form.
- Run one command
Helm on a cluster, or one line on each Linux server. A deb, an rpm and docker run work too.
Kubernetes
helm install watchlog-agent … --set watchlog.clusterToken=wl_ct_••••Linux server
curl -fsSL https://ingest.watchlog.io/install.sh | sudo sh -s -- --token wl_ct_•••• - Watch it connect, live
A checklist ticks as the agents, the metrics and the service map arrive. First data in under a minute in our tests.
Works withKubernetes
Docker
Linux servers
and reads
Redis
PostgreSQLMySQL and MariaDB
MongoDB
NginxTraefik
RabbitMQKafka
PHP-FPM

Your clusters’ and servers’ data, kept like it matters.
The agent is read-only and never reads Secrets, ConfigMap data or environment values. In the Cloud, every organization is isolated, and that isolation is tested on every change.
- On a server: its own user, not root
- Five capabilities, no more
- Command lines and environments never leave the server
- What it reads →
Isolated per organization
Row-Level Security in the database and an organization check on every query, API call and MCP tool, tested on every change.
How Cloud keeps data apart →A DPA and our subprocessors
Read the data processing agreement and the full list of services that process data for Cloud. Backups are encrypted before they leave the server.
Subprocessors →2FA and audited access
Two-factor sign-in for any account, required for Owners of paid organizations. Staff see your data only in a read-only support session you can review.
Support sessions →Leave any time
Export your configuration and metrics, or delete the organization, with 7 days to undo. Or move to the free self-hosted edition: same product.
Export and deletion →
The answers you need, without assembling a stack.
Why is this broken? WatchLog tells you.
Every alert, failing connection and crashing pod, container, service or process gets its probable cause: one sentence, the evidence behind it, and what was checked. On a server too: a pm2 app that crash-loops after a release, a service OOM-killed at its MemoryMax, PHP-FPM at pm.max_children. Fixed rules over the data WatchLog already collects; no AI, no scores.
frontend fails because api was scaled to 0: nothing answers its calls.
shop-worker-1 is OOM-killed when it reaches its 256 MiB memory limit: 12 times in the last hour.
api exits with code 1 since its executable changed at 14:01: a new release, 1 min before the first crash.
orders.service is OOM-killed at its 512 MiB memory limit: 3 times in the last hour.
Every worker was busy 14 times in the last 10 minutes (pm.max_children reached); 6 requests are waiting.


What changed: anomaly detection
Each metric learns its usual week. When something breaks, one list shows what left its range first, with the probable cause underneath. In Cloud Pro.
How What changed works →Service map from real traffic
eBPF sees every connection: who talks to whom, failures and latency, between pods, containers and the processes on a server. No sidecars, no code changes.
What eBPF reads →Egress watch: what talks to the internet
Every external destination of every workload, container or server process, since when, and an alert when one starts talking somewhere new. From connections only: no payload read.
How Egress watch works →Restart reasons, not just counts
OOMKilled, exit codes and crash loops explained in plain words, for pods, containers, systemd services and processes, with the change that came before.
See it in the live demo →Alerts that work on day one
Sensible rules switched on at install, to Slack, Telegram, email or a webhook, with the likely cause in the message.
The default rules →The services next to your apps
Redis, PostgreSQL, MySQL and MariaDB, MongoDB, Nginx, Traefik, RabbitMQ, Kafka and PHP-FPM found automatically, in a cluster, in Docker or installed on the host, each with its own page.
Applications →Logs on demand
The last lines of a pod's or a container's log, a service's journal or a pm2 app's log, where you already look: redacted, streamed and never stored.
How logs work →Right-sizing hints
Requests and limits, and a service’s MemoryMax and CPUQuota, against seven days of real use, in plain sentences: what was set, what was used, what would cover it. No scores.
What each hint means →Press ⌘K, go anywhere
Pages, clusters, workloads, pods, nodes, servers, services, containers and applications, found by name with ⌘K or Ctrl+K. Keyboard first.
How to find anything →Linux servers: systemd services, process groups and containers
Every unit, every process in any language and every container on a server, with crashes and OOM kills from the kernel and, new in 0.9.2, each server’s own events: failed, restarted, crash-looping, released. One command per server.
Linux server monitoring →
Ask your AI what broke. It answers from WatchLog.
Connect Claude or ChatGPT with a read-only token. It reads real exit codes, OOM kills, slow queries and rollouts, on Kubernetes, Docker and Linux servers: a pod, a container, a systemd service or a pm2 app.
Try it on the live demo →How it works
Read only: tokens can't change anything and never see secrets, and every call is logged. No AI runs inside WatchLog.
What's wrong with my clusters?
Asking WatchLog…
- get_problems182 ms2 critical problems in shop
- explain_alert95 msimage-resizer · OOMKilled, exit 137
- explain_alert88 mspayments-api · exit 1, CrashLoopBackOff
- get_events61 msWarning events of the last 2 h
- get_service_map112 msno failing connections
kind-demo is red because two pods in shop keep crashing.
1.shop/image-resizer goes over its 32Mi memory limit and gets killed (OOMKilled, exit 137). Nothing was deployed before it.
2.shop/payments-api exits with code 1 and is in CrashLoopBackOff. It isn't memory: read its last run with kubectl logs --previous.
No nodes are down, and no connections are failing.
Recorded with Claude Code against WatchLog, shortened. Every MCP client gets the same tools.
What's wrong with my clusters?
Asking WatchLog…
- get_problems140 ms2 critical, 4 warnings
- list_alerts61 ms3 alerts firing
- explain_alert83 msstorefront-api · exit 2, restarting
- get_pod77 msbilling-worker · OOMKilled, exit 137
- get_node52 msdocker-02 · silent for 7 days
prod-fleet has a silent host and two containers in trouble.
1.docker-02 stopped reporting 7 days ago. Check that the host and its WatchLog agent are up, or remove it.
2.billing-worker was killed at its memory limit (OOMKilled, exit 137), three times so far.
3.storefront-api exits with code 2 every few seconds and Docker starts it again: 35 restarts. Read docker logs storefront-api.
Recorded with Claude Code against WatchLog, shortened. Every MCP client gets the same tools.
What's wrong with my servers?
Asking WatchLog…
- get_problems1 critical, 1 warning on web-2
- get_serverweb-2 · services, process groups, containers
- get_eventsapi · CrashLoop, Release, FailedExit
web-fleet has two workloads in trouble, both on web-2.
1.api a pm2 app, exits with code 1 since its executable changed at 14:01: a new release, 1 min before the first crash.
2.orders.service is OOM-killed at its 512 MiB memory limit: 3 times in the last hour. systemd restarts it each time.
The other two servers are healthy.
An example on a Server fleet, written from the tools’ docs. Every MCP client gets the same tools.
Your databases, caches and queues, too.
Found automatically in what the agents already see: in a Kubernetes cluster, in Docker, or installed on the host of a Linux server. Connect with the command WatchLog writes for you, and each gets a full page, alerts and plain-language reasons.
Redisand Valkey
PostgreSQLqueries, locks, replicationMySQL and MariaDBconnections, lock waits, replication
MongoDBoperations, replica set, slow scans
Nginxrequests, 5xx, latencyTraefikrequests, 5xx, backends down
RabbitMQqueues and consumersKafkapartitions, consumer lag
PHP-FPMpools, busy workers, pm.max_children
ElasticsearchPlanned
Simple pricing: three dollars per node or server.
Billed on the 95th percentile of hourly node counts, so autoscaling spikes don’t inflate the bill. A Linux server counts as one node.
Cloud Free
$0forever
Start free- 1 cluster or Server fleet
- Up to 5 nodes or servers
- 3 days of data
- 1 user
- Alerts, service map, probable cause, MCP
Cloud Pro
$3/ node or server / month
$15 minimum, the first 5 · annual: 2 months free
Start with Pro- What changed and anomaly alertsNEW
- Unlimited clusters, nodes and servers
- 30 days of data
- Up to 10 users
- Email support, within 2 business days
Coming to Pro: the multi-cluster view and SLO reports
Community
Freeself-hosted
Install it yourself- Unlimited clusters, fleets, nodes and servers
- History you choose (15 days by default)
- Unlimited users: Owner, Admin, Viewer
- AGPL-3.0, one Helm chart or Compose file
Nothing is charged without a card you add on Stripe’s own page. Cloud Free and the self-hosted Community edition stay free: nothing in them will become paid. Organizations that started Pro during the beta keep it free until the date in their notice, sent 14 days ahead.

