Skip to main content

Goal

Register a webhook endpoint so your backend can react to Sippet activity in near real time.

Before you start

  • You need a reachable backend endpoint that can accept webhook requests.
  • Decide which support events matter to downstream systems first.

Step 1: create the endpoint

In Settings > Webhooks:
  • add the target URL
  • verify the endpoint belongs to the correct environment
  • keep the first subscription small and observable

Step 2: decide what happens downstream

Common webhook consumers include:
  • internal support orchestration services
  • analytics pipelines
  • CRM or ticketing sync jobs
Common event families include:
  • contact.*
  • call.*
  • conversation.*
  • communication.message.*
  • email.classification.*
  • callback.*
  • playbook.run.*

Email classification payload

The email classification events use the standard webhook envelope with an email classification record in data.
predicted_label always preserves the model choice. When confidence is below confidence_threshold, effective_label is unclassified. Manual override and clear updates include before_label and after_label so consumers can record the transition. The initial successful lifecycle emits email.classification.created followed by email.classification.classified. The first transition into the classified state does not also emit email.classification.updated; that event is reserved for later state changes, reclassifications, and manual label updates.

Step 3: monitor delivery history

After enabling the endpoint:
  • review delivery results
  • investigate repeated failures quickly
  • confirm retries do not create duplicate downstream work

Example workflow

  • a support event occurs in Sippet
  • your webhook endpoint receives the event
  • your backend stores, routes, or transforms the event for another system

What good looks like

  • deliveries succeed consistently
  • your downstream system stays in sync
  • failures are visible and actionable from the console

Troubleshooting

  • If deliveries fail, start by checking endpoint availability and response behavior.
  • If your downstream job is too broad, narrow the events you act on first.