Send Prometheus Alertmanager alerts to WhatsApp
A webhook receiver in alertmanager.yml and a key. No
phone number of your own, no Meta business account, no template review, and no relay
service of yours to keep alive.
Alertmanager has no WhatsApp receiver, and the usual answers to that are a paid
messaging account with per-conversation pricing, or a small gateway you write that
holds a personal WhatsApp number open as a linked device — the kind that works
until the number gets restricted. This route is a webhook_configs block
pointed at an endpoint that already speaks Alertmanager's payload, delivering over
Meta's official Cloud API on our registered number.
It takes about two minutes, and the first three alerts are free.
1. Make a key
In the Jabee dashboard, go to Management then
Approvals and create an API key. It starts jak_. Copy it
now; it is not shown again.
Whoever is going to be woken up verifies a phone by tapping a link in the dashboard and sending one message. Until somebody has done that there is nobody to deliver to.
2. Add the receiver
In alertmanager.yml:
receivers:
- name: whatsapp
webhook_configs:
- url: https://api.jabee.ai/api/v1/alerts/grafana
send_resolved: false
http_config:
authorization:
type: Bearer
credentials: jak_your_key_here
The endpoint path says grafana because Grafana's webhook body is
Alertmanager's, extended; both post to the same place and both are read natively.
There is no notification template to write — the default payload is the
point.
Set send_resolved deliberately. Alertmanager
defaults it to true, and this end drops recoveries unless the URL opts
in — harmless, and a lot of pointless requests. Either set it
false as above, or leave it true and add
?resolved=true to the URL if you want the all-clear message. A dropped
recovery answers 200 OK with {"ignored": "resolved"}
rather than an error, because Alertmanager retries anything it reads as a failed
delivery.
If the key should not live in the config file, credentials_file
takes a path instead:
http_config:
authorization:
type: Bearer
credentials_file: /etc/alertmanager/jabee_key
3. Route to it
A receiver does nothing until the routing tree sends something to it. The three timers on the route decide how often a continuing problem messages you, and on WhatsApp that is the difference between being informed and being nagged:
route:
receiver: whatsapp
group_by: ["alertname", "cluster"]
group_wait: 30s # let a bad minute collect into one message
group_interval: 5m # a changed group waits this long to post again
repeat_interval: 4h # a continuing incident reminds you this often
Or keep your existing tree and route only what should wake a person:
route:
receiver: default
routes:
- matchers: ["severity = critical"]
receiver: whatsapp
repeat_interval: 4h
repeat_interval is the one to set consciously: every repeat is a
real message against your monthly allowance. Four hours is a reasonable start for
something that pages a human; the default of four hours only applies if you have not
inherited something shorter higher up the tree.
4. Fire a test alert
No need to break production. Post a synthetic alert straight to Alertmanager's API and watch the phone:
curl -XPOST http://localhost:9093/api/v2/alerts \
-H 'Content-Type: application/json' \
-d '[{
"labels": {"alertname": "JabeeWebhookTest", "severity": "critical"},
"annotations": {"summary": "Test: Alertmanager can reach WhatsApp"}
}]'
After group_wait passes — 30 seconds in the config above
— a WhatsApp message arrives. If it does not, the checklist is short: is a
phone verified in the dashboard, is the key on the Authorization header
as Bearer jak_..., and did the test alert actually match the route that
points at the whatsapp receiver (amtool config routes test
answers that one).
What arrives
Alertmanager groups alerts by your group_by labels, and a group
arrives as one WhatsApp message, not one per alert: the first few
alerts get a line each with their labels, the rest are counted, and the message
carries the generatorURL's origin so you can get to the detail. One
notification is one alert against the allowance however many problems it carries, so
sensible grouping makes the 500 go a long way.
Kubernetes
On kube-prometheus-stack, the same receiver and route go under
alertmanager.config in your Helm values (or an
AlertmanagerConfig resource if you manage routing that way). Nothing
about the endpoint changes; the only Kubernetes-specific advice is to mount the key
from a Secret and reference it with credentials_file rather than
inlining it in values that end up in a git repository.
One honest caveat: no deduplication
If Alertmanager retries a delivery it never saw succeed, you may get the same
message twice. The obvious key to deduplicate on is the group key, and it is the
wrong key: it is reused for every repeat of an ongoing incident, so suppressing on
it would silence exactly the reminders repeat_interval exists to send.
A duplicate message is a smaller harm than an alert that never arrives, so that is
the trade made here.
What it costs
$29 a month for 500 alerts and approvals together, then 3 cents each, with no per-seat charge. The first three are free, so you can add the receiver, fire the test alert above, and watch one arrive before paying for anything.
Alerts go to one verified phone per business — the account owner's, or an ops handset verified in the dashboard under Where alerts go. To route different alerts to different people, connect a WhatsApp number of your own and send from there.
Questions
Does Alertmanager have a native WhatsApp receiver?
No. The shipped receivers cover email, Slack, PagerDuty, OpsGenie, webhooks and
a few others; WhatsApp is not among them, which is why every route to it goes
through webhook_configs one way or another. The difference between
the routes is what sits behind the webhook: a paid messaging account, a gateway
holding a personal number open, or an endpoint like this one on Meta's official
Cloud API with nothing of yours attached.
Do I need a WhatsApp Business account or my own number?
No, neither. The number and the Meta-approved template are ours, so you need an API key and nothing else. There is no number of yours involved, so there is nothing of yours to get banned.
Does this work with plain Prometheus, VictoriaMetrics or Mimir?
Yes, wherever the alerts end up in an Alertmanager-compatible process. The receiver reads Alertmanager's webhook payload, so anything that emits that shape — vmalert pointed at Alertmanager, Mimir's ruler, Grafana-managed rules — lands the same way.
Why did my test send nothing?
Three usual causes, in order: no phone verified yet (open the dashboard, tap
the link, send the one message it writes for you); the key missing from the
Authorization header or missing its Bearer scheme; and a
routing tree that never sends the test alert to the receiver —
amtool config routes test with the test alert's labels shows which
receiver would fire.
Can different severities page different people?
Not on Jabee's number — alerts go to the one phone this business
verified, so there is nothing to address. Route critical to WhatsApp
and everything else to email or Slack, or connect your own WhatsApp number and
route per person from there.