First-in-first-out, per endpoint
Each endpoint gets its own Durable Object, which preserves the order events arrived in and isolates one endpoint’s backlog from another’s. A slow or failing destination slows only itself.
delivery
Received once, in order, never silently dropped
The same engine that captures your events runs the pipeline that moves them onward. Each endpoint keeps first-in-first-out order and its own isolation; failed deliveries retry, and what still can't land is dead-lettered, not dropped.
event twilio · message.received
happy path
receiveverifydedupqueuedeliver
on failure
retrybackoffdead-letterreplay
first-in-first-out per endpoint · held, not dropped
Retries with backoff, then a dead letter
A delivery is attempted up to eight times over roughly 28 hours — the first try, then seven retries with growing gaps: 5 seconds, 5 minutes, 30 minutes, on out to ten-hour waits. What still can’t land is held in a dead-letter queue, not dropped, so you can fix the destination and replay. An endpoint that fails 20 times in a row is auto-disabled, so we stop hammering a target that’s clearly down — and tell you.
Deduplicate by id, when you want it
Providers resend. Turn on idempotency-key deduplication per endpoint and a repeated event is recognised and dropped before it reaches your handler twice. It’s off by default — opt in per endpoint, because what counts as a duplicate is your call, not ours.
Point a webhook at it. Watch it land.
Start free: a permanent URL, full inspection, one-command replay — and outbound delivery, because every feature is on every plan. Move up when you need more events.