Issue No. 04 — On delivery guarantees
New · Event replay from the CLI
Most teams build webhook delivery twice: once in a sprint, and again after the first outage. This is the version you build zero times.
Sending events in production at
How it works
No queue to run, no cron jobs, no retry table to migrate.
Install the SDK and call send() wherever your app emits an event. Existing JSON payloads work as they are.
Pick a region, a retry policy and rate limits per endpoint. Defaults are sensible; every knob is optional.
Each delivery shows up in the log with status, latency and the response body. Replay anything that failed.
Under the hood
01 SDKs
Pass an event type, a payload and an idempotency key. The SDK handles auth, signing and timeouts, and returns a message ID you can look up later. Switch tabs and the call looks the same in every language.
02 Delivery log
The log lists each delivery attempt with its HTTP status, latency and response body. Filter for 5xx responses from one endpoint, see that the customer's server timed out at 30 seconds, and replay the batch once it is back.
03 Regions
Six regions, each with its own delivery workers. The latency chart shows p50 and p95 per region, measured from our edge to the receiving server, so you can tell a slow customer apart from a slow network.
Features
The boring, load-bearing parts of event delivery, built once and run for you.
Up to 12 attempts over 72 hours with jittered exponential backoff. Failed events land in a dead-letter queue, never on the floor.
Every request carries an HMAC-SHA256 signature and timestamp. Rotate signing secrets with a 24-hour overlap and zero downtime.
Search 30 days of deliveries by event type, endpoint or status code. Inspect the exact request and response, then replay with one click.
Node, Python, Go, Ruby, PHP, Java, .NET and Rust. Event schemas generate types, so a renamed field fails your build, not your customer.
Send to us-east, us-west, eu-west, eu-central, ap-southeast or sa-east. Payloads stay in the region you pick.
Cap delivery to 50 requests a second for a fragile customer server. We queue the rest and keep ordering per key.
Testimonials
“Signature verification took me ten minutes in Go. Our old setup had a shared secret nobody remembered how to rotate.”
“A customer's endpoint was down for six hours over a weekend. Every event arrived when it came back, in order. I found out on Monday from the log.”
“Support can search the delivery log by customer ID and replay events themselves. That used to be an engineering ticket every day.”
“The typed Python SDK caught a renamed field in CI before it reached 1,200 integrations. That alone paid for the year.”
Pricing
Retries, replays and signature checks are always free. You only pay for events you send.
| Compare plans | Hobby$0per monthFor side projects and prototypes.Start free | ScaleMost popular$49per monthFor production apps with real customers.Choose Scale | EnterpriseCustomannualFor regulated teams and high volume.Talk to sales |
|---|---|---|---|
| Events per month | 50k | 1M + $20/M | Custom |
| Endpoints | 3 | Unlimited | Unlimited |
| Delivery log | 7 days | 30 days | 90 days |
| Regions | us-east | All 6 | All 6 + dedicated |
| Event replay | |||
| Uptime SLA | 99.99% | 99.99% + credits | |
| SSO and SCIM | |||
| Support | Community | Email, 1 business day | Slack, 1 hour |
Prices in USD, billed monthly. Overage on Scale is $20 per additional million events. Cancel any time.
FAQ
Something missing? Email us; an engineer answers, usually within a few hours.
Contact supportOne API call to send, regardless of how many endpoints it fans out to or how many retries it takes. An event delivered to five subscribers and retried twice is still one event on your bill.
Any non-2xx response or a timeout after 30 seconds triggers a retry. The default schedule is 12 attempts over 72 hours with jittered exponential backoff. You can shorten it per endpoint, and after the last attempt the event moves to a dead-letter queue you can replay.
Each request includes a signature header with an HMAC-SHA256 of the timestamp and body. Every SDK ships a verify() helper, and the docs include copy-paste snippets for frameworks that don't have an SDK yet.
Payloads are stored in the region you choose when you create a workspace and are deleted when your log retention ends: 7, 30 or 90 days depending on plan. We never move data between regions.
Most teams dual-send for a week: keep the old worker running, send the same events through Dispatchly with delivery paused, compare the logs, then flip endpoints over one customer at a time. Our team will review your plan for free.
We never drop or throttle events because of billing. On Hobby we email you at 80% and 100% and ask you to upgrade; on Scale additional events are billed at $20 per million at the end of the month.
The 99.99% SLA covers the send API accepting events and our workers attempting delivery. If we miss it in a calendar month, you get service credits of 10 to 50% of that month's bill, applied automatically.
Yes. Downgrade to Hobby or close the workspace from settings. Paid plans are billed monthly with no annual commitment, and you can export your delivery log as JSON before you go.