Configuration

createMonitor

const monitor = createMonitor<Env>({
  d1: (env) => env.MONITORING_DB,
  secrets: (env) => ({ SLACK_WEBHOOK_URL: env.SLACK_WEBHOOK_URL }),
  baseUrl: (env) => env.BASE_URL,
  auth: (env) => ({ teamDomain: env.CF_ACCESS_TEAM_DOMAIN, aud: env.CF_ACCESS_AUD }),
  notifiers: [slack({ webhook: '${SLACK_WEBHOOK_URL}' })],
  headerRules: [],
  resolveUrl: (url, env) => url,
  defaults: {},
});

Option	Required	Purpose
d1	yes	The D1 binding holding checks, state, and history
secrets	no	Values that checks and notifiers reference as ${NAME}
baseUrl	no	Public URL, used to build the links inside alerts
auth	no	Access configuration. Without it, writes are refused
notifiers	no	Where alerts go. Defaults to Slack if SLACK_WEBHOOK_URL is set
headerRules	no	Headers applied to every check on a matching host
resolveUrl	no	Rewrites check URLs at run time, e.g.
defaults	no	Fallbacks for check options and run behaviour

defaults

Key	Default	Notes
failureThreshold	3	Consecutive failures before a check is unhealthy
timeoutMs	10000	Per-request timeout
retryCount	1	Retries on failure only; a passing check never retries
retryDelayMs	5000	Wait between retries
reminderIntervalMs	3600000	Re-alert cadence while down. null disables
intervalMins	5	How often a check runs
concurrency	6	Checks executed at once
historyRetentionHours	48	How long results are kept
userAgent	clawdwatch/3.0	Sent unless a check overrides it
captureBodyOnFailure	false	Keep a scrubbed excerpt of a failing response. See AI agents

Check options

{
  "id": "api-health",
  "name": "API Health",
  "url": "https://api.example.com/health",
  "method": "POST",
  "headers": { "Content-Type": "application/json" },
  "body": "{\"query\":\"{__typename}\"}",
  "assertions": [{ "type": "statusCode", "operator": "is", "value": 200 }],
  "retryCount": 1,
  "retryDelayMs": 5000,
  "timeoutMs": 10000,
  "failureThreshold": 3,
  "reminderIntervalMs": 3600000,
  "intervalMins": 5,
  "tags": ["production", "api"],
  "enabled": true,
  "captureBodyOnFailure": false
}

id may contain letters, digits, ., _, and -. Everything except id, name, and url is optional.

Intervals and the cron

intervalMins cannot make a check run more often than your cron trigger fires. With the default */5 * * * *, a check asking for one minute still runs every five.

Scheduling allows half an interval of slack, so a cron firing a few seconds early does not skip a tick and silently turn a five-minute check into a ten-minute one.

Assertions

Every assertion must pass. Failures are reported together, so one run tells you everything that was wrong.

statusCode

{ "type": "statusCode", "operator": "is", "value": 200 }

Operators: is, isNot.

header

{ "type": "header", "name": "content-type", "operator": "contains", "value": "json" }

Operators: is, isNot, contains, notContains, matches.

body

{ "type": "body", "operator": "contains", "value": "\"healthy\"" }

Operators: contains, notContains, matches.

responseTime

{ "type": "responseTime", "operator": "lessThan", "value": 3000 }

Measured from request start to response headers.

jsonPath

{ "type": "jsonPath", "path": "$.data.status", "operator": "is", "value": "ok" }

Operators: is, isNot, contains, notContains, matches, lessThan, greaterThan.

Path syntax is deliberately small: $.field, $.nested.field, $.array[0].field. No wildcards, no recursive descent. A missing path fails the assertion. Non-string values are stringified before comparison, so $.count compares against "42".

Use isNot with an empty string to assert a field is present and non-empty:

{ "type": "jsonPath", "path": "$.data.token", "operator": "isNot", "value": "" }

Maintenance windows

A window suppresses alerts, skips checks entirely, or both. It can target one check, a tag, or everything.

INSERT INTO maintenance_windows
  (id, check_id, tag, starts_at, ends_at, reason, suppress_alerts, skip_checks)
VALUES
  ('deploy-friday', NULL, 'production', '2026-08-01T22:00:00Z',
   '2026-08-01T23:30:00Z', 'Planned deploy', 1, 0);

With suppress_alerts, checks still run and history still records the outage — you just are not paged for it. That is usually what you want: the dashboard should still show what happened.

Config as code

Export every check, with secret references intact, and commit the result:

curl "https://your-worker.workers.dev/api/config" > checks.json

Apply it from CI:

curl -X PUT "https://your-worker.workers.dev/api/config" \
  -H 'Content-Type: application/json' \
  --data @checks.json

The UI and this file write through the same validation, so you can use either or both.