Calling it
When you enable the workflow, the app shows its Webhook URL and a token, once. Either way in is enough:- Token in the path:
POST <your Reliant URL>/hooks/<trigger id>/<token>. This is for senders that cannot set headers, such as Zapier. Treat the whole URL like the token. - Bearer token:
POST <your Reliant URL>/hooks/<trigger id>withAuthorization: Bearer <token>.
header defaults to X-Signature-256, algorithm to sha256 (sha1 and sha512 also work), encoding to hex (or base64). Configuring HMAC adds a way in; the token still works.
The server needs the shared secret when the automation is enabled. Give it to reliant trigger activate from a file, keyed by the trigger’s name (- reads it from standard input):
activate again with a new file to replace it. Without one, enabling a webhook that declares hmac is refused with webhook_hmac_secret is required to verify signatures. The app’s Enable dialog does not take a secret yet; until it does, enable a signed webhook from the CLI, or use the token.
The token is stored hashed and shown once: reliant trigger activate prints it with the URL when it creates the activation. If it leaks, rotate it with reliant trigger rotate <name>, or from the automation’s page: the old token stops working immediately, and the new one is printed once.
Your organization’s alerts
A webhook declared withsource: alerting is delivered to by Reliant’s own alerting rather than by a sender you configure:
webhook_config at it with the token shown at Enable. The built-in Autopilot on-call workflow declares its trigger this way.
What the run sees
trigger.payload has four fields: body (the parsed JSON, or the raw text if it is not JSON), headers, query and content_type. Headers and query entries that carry secrets are removed. Payloads are untrusted data from an outside sender, so guard optional fields with has() in filters.
Responses, retries and duplicates
By default the
202 body contains event_id, outcome, and duplicate. The event has been recorded and evaluated, but a run may not have started.
Send Prefer: respond-async if you only need to know the delivery was stored and want filtering and dispatch to happen in the background. A 202 with Preference-Applied: respond-async returns {"delivery_id":"…","status":"pending"}. The delivery ID is a receipt, not a trigger event ID; check the automation’s history for the eventual result.
For either mode, the first non-empty header among Idempotency-Key, X-Idempotency-Key, and X-Delivery-Id identifies the delivery. X-Request-Id is ignored for deduplication: it identifies a request, not an event, and proxies can assign a new value on each retry.
Without an idempotency header, the default mode treats an identical body within the same minute as a retry. Async mode stores each POST as a separate delivery, even when the bodies match. To deduplicate retries in either mode even when the body changes, send the same Idempotency-Key on each attempt or declare a workflow dedupe_key (below).
Deduplicating by what the event is about
A sender with no idempotency header, such as Alertmanager, re-sends the same alert group on everyrepeat_interval, with a body that has changed, so the default key counts each repeat as a new event. A workflow that declares the webhook can name the event itself:
dedupe_key is a template over trigger, the same root a filter reads. It replaces the header-or-body default for this trigger. Deliveries that render the same key within one dedupe_window are one event: the first starts a run, and the rest are recorded in the automation’s history as duplicates and start nothing. Here a re-sent firing notification for a group is one run, while its resolved notification (a different key) is another.
The window is explicit:
- Default 1 hour, at most 24 hours (at least one second). Write it as a duration:
30m,4h. - It is a fixed clock bucket aligned to UTC, not a sliding window. Two deliveries that straddle a bucket boundary are two events, so the window is the longest a repeat can be suppressed, and may be shorter. An alert that is still firing after the window is reported again, which is usually what you want.
- It needs a
dedupe_key, and applies to webhook triggers only. Integration, schedule and workflow-event triggers already dedupe by their own event id.
groupKey, or it renders empty) falls back to the default key for that delivery, so a key that does not fit the payload never drops or merges events. Save-time validation checks that the template compiles, has a {{ }}, and that the window is a duration in range. To see what a key renders to before you enable anything:
Testing
Send a request withcurl, then read the result in the automation’s history, which records each delivery and what it did:

A webhook delivery that started a run. Example data.
Related topics
- Automations overview
- HTTP integration, for calling out rather than in