Website down alerts for AI agents and scripts
Bionic Uptime can post each confirmed outage and recovery to a URL of yours, as signed JSON. An AI agent, an n8n or Zapier workflow, or a few lines of your own code can then look into it, open a ticket or tell the client. It is in addition to the email, never instead of it, and it is on every plan, free included.
Why confirmation matters more for an agent
A person who gets a false alarm sighs and checks. An agent that gets one goes and "fixes" a website that was never broken: it restarts a server, rolls back a deploy or emails the client. So we post only what we would email: an outage is sent only after two locations on two different companies' networks, Chicago and New Jersey, and a retry all agree inside 45 seconds. A single slow answer from one place is never posted. How we avoid false alarms.
1. Turn it on
Sign in and open your account page. Under "Webhook for AI agents and scripts", paste an https URL that answers a POST with any 2xx status. We post a signed test event before saving, so you know it works. The page then shows your signing secret, which starts with whsec_.
2. What we send
One POST per event, with Content-Type: application/json. type is website.down when an outage is confirmed, website.up when the website is confirmed back, or test from the account page. This is a real website.down body, for a made-up website:
{
"type": "website.down",
"timestamp": "2026-09-15T18:02:33Z",
"data": {
"summary": "Riverside Landscaping is down (riversidelandscaping.example). Confirmed from both checking locations at 2:02 PM.",
"website": {
"id": "6f1c2b4e-0000-4000-8000-000000000001",
"name": "Riverside Landscaping",
"url": "https://riversidelandscaping.example/"
},
"incident": {
"id": "9a7d3c10-0000-4000-8000-000000000002",
"state": "confirmed",
"started_at": "2026-09-15T18:02:10Z",
"confirmed_at": "2026-09-15T18:02:31Z",
"error_category": "timeout",
"failure": "Both checking locations waited ten seconds without a response.",
"url": "https://bionicuptime.com/websites/6f1c2b4e-…?incident=incident-9a7d3c10-…"
},
"evidence": [
{ "at": "2026-09-15T18:02:10Z", "location": "chicago", "label": "Chicago",
"kind": "routine", "ok": false, "category": "timeout",
"detail": "No response within 10 seconds", "duration_ms": 10000 },
{ "at": "2026-09-15T18:02:20Z", "location": "newjersey", "label": "New Jersey",
"kind": "crosscheck", "ok": false, "category": "timeout",
"detail": "No response within 10 seconds", "duration_ms": 10000 },
{ "at": "2026-09-15T18:02:31Z", "location": "chicago", "label": "Chicago",
"kind": "retry", "ok": false, "category": "timeout",
"detail": "No response within 10 seconds", "duration_ms": 10000 }
],
"confirmed_from": ["chicago", "newjersey"]
}
}- evidence is every check the outage was confirmed from: where (
locationischicagoornewjersey), when, what kind of check, and what it saw. Awebsite.upevent adds the checks that confirmed the recovery, and hasrecovered_at. - incident.url opens the incident in Bionic Uptime. It needs you signed in, so it is safe to pass on to an agent.
- Fields may be added later. None will be renamed or removed.
3. Check it is really us
Each post carries three headers in the Standard Webhooks format: webhook-id, webhook-timestamp and webhook-signature. The signature is an HMAC-SHA256 of the id, the timestamp and the body, joined by dots, keyed with your secret. Their free libraries check it in one line. Python (pip install standardwebhooks):
from standardwebhooks.webhooks import Webhook
wh = Webhook("whsec_...") # your signing secret
event = wh.verify(raw_body, headers) # raises if it is not from usNode (npm install standardwebhooks):
import { Webhook } from "standardwebhooks";
const wh = new Webhook("whsec_..."); // your signing secret
const event = wh.verify(rawBody, headers); // throws if it is not from usCheck the raw body, before any JSON parsing changes its spacing. If the secret leaks, make a new one on the account page; events from then on are signed with it.
4. Answer fast, and expect a repeat
- Reply 2xx within ten seconds, then do the slow work, such as starting an agent. A receiver that takes longer looks unreachable.
- We retry on a timeout, a 429 or a 5xx, waiting longer each time. A 400, 401, 403, 404 or 410 is taken to mean the URL is wrong: we stop, and the account page says so.
- Drop repeats by webhook-id. A retry carries the same id. If your reply to us is lost on the way back, we post again, because a real outage we never re-sent would be worse.
- Down, then up. If an outage could not be delivered before the website came back, the down event is dropped and only the up event goes, so the last thing your agent hears about a website is never "down" when it is up.
5. Recipes
- n8n. Add a Webhook node set to POST, copy its production URL into the account page, and branch on
body.type. Point an AI Agent node atdata.summary,data.evidenceand the website's URL. - Zapier or Make. Use their "catch a webhook" trigger and paste its URL in the same way.
- An agent on your own server. A small receiver checks the signature, answers 200, and then hands the event to whatever agent you run, for example with Claude Code:
claude -p "This website is down. Find out why and draft a note to the client. Do not change anything." < event.json. Ask it to look and report first; let it act only on what you trust it with.
What it will not do
It posts to one URL per account, over https on the standard port, to a public address. Private, local and cloud metadata addresses are refused, and we do not follow redirects. It does not post "Status uncertain", certificate or domain reminders; those stay email only for now. And it never replaces the email: if your receiver is down, the alert still reaches the people on that website's list.
3 websites are free, forever, with no card.