Observability

Effect already emits traces, metrics, and logs. Alchemy declares the exporter as a Layer — point it at Axiom, Datadog, CloudWatch, or any OTLP endpoint. Then declare your dashboards and alarms next to the code that emits the metrics.

OTel everywhere

Every Effect spans a trace. Every Metric increments a counter. Every logInfo ships a log line. The exporter is a Layer — swap one line, ship to a different vendor.

// Effect emits OpenTelemetry by default.
// Pick an exporter Layer; the Worker code never changes.
export default class Api extends Cloudflare.Worker<Api>()(
  "Api",
  Effect.gen(function* () {
    yield* Effect.logInfo("request received");
    yield* Metric.increment(requestsTotal);
    return { fetch: handler };
  }).pipe(
    Effect.provide(AxiomExporter),       // or CloudWatch, Datadog, OTLP …
  ),
) {}

Supported destinations include Axiom, Datadog, CloudWatch, and any OTLP-compatible collector.

Dashboards and alarms in code

Declare CloudWatch (or Grafana, or Datadog) dashboards and alarms next to the resources they observe. Threshold change? It’s a diff in the PR — reviewable, revertable, audited.

// alchemy.run.ts — same program. same diff. operations included.
export const Dashboard = AWS.CloudWatch.Dashboard("ApiHealth", {
  widgets: [
    Widget.line({   title: "p99 latency",     metric: api.metrics.p99 }),
    Widget.line({   title: "requests / sec",  metric: api.metrics.rps }),
    Widget.number({ title: "5xx ratio",       metric: api.metrics.errorRate }),
  ],
});

export const P99Alarm = AWS.CloudWatch.Alarm("p99Latency", {
  metric: api.metrics.p99,
  threshold: 500,
  comparisonOperator: ">",
  evaluationPeriods: 5,
  alarmActions: [pagerDuty, slackWebhook],
});

• Widgets reference typed .metrics outputs from your resources.
• Alarms compose. Same alarm wired to PagerDuty, Slack, SQS — typed actions.
• Per-stage dashboards: prod, staging, pr-42 all separate.

Alerts wired to anything

Alarm actions are typed resources too. Wire your alarms to PagerDuty, Slack, an SQS queue, an SNS topic — reuse the same notification channel across every alarm in the stack.

• Threshold + window — Configure evaluation periods, statistic, comparison operator — all typed, autocompleted in your editor.
• Composite alarms — AND/OR multiple alarm states into a higher-level alert — declared as values, not console rules.
• Audit trail in git — Every threshold and channel lives in source. Who changed it, when, and why — answered by git blame.