What model does the classify step use?

Yours. You bring the model, prompt, and label schema; ingestlayer runs it in flight and returns a typed label the pipeline branches on.

Does every ticket hit the model?

Only if you want it to. classify is per-event and cached by payload hash, so identical tickets reuse one call.

Can the same ticket go to two places?

Yes — fan out to several destinations with different when conditions, e.g. critical to chat and everything to Postgres.

ingestlayer/recipes/v0.4

Route support escalations in Telegram

Classify inbound tickets by urgency in flight and route only the ones that need a human now — so the on-call channel sees escalations, not every ticket.

01source

sourcesdk.eventTypeScript SDK

matchsupport.ticket.created

02pipeline · 3 steps

01ENRclassifyurgency: low | high | critical
02CTLfilter.matchurgency = critical only
03MUTredact.piistrip PII from body before posting

03destinations · 1

totelegramTelegram
chat@oncall

the event

You emit support.ticket.created with this shape. The TypeScript SDK keeps the call type-safe, and the event is stored whole — so every field below is available to the pipeline by name.

ticket_idstring
subjectstring
bodystringfree text
customer_tierstringfree | pro | enterprise
channelstringemail | chat | form

emit it

From your code with the TypeScript SDK — or any language over the REST endpoint and signed webhook ingress.

emit support.ticket.created

import { ingest } from "@ingestlayer/sdk";

await ingest("support.ticket.created", {
  ticket_id:     ticket.id,
  subject:       ticket.subject,
  body:          ticket.body,
  customer_tier: ticket.account.tier,
  channel:       ticket.channel,
}, {
  idempotencyKey: ticket.id,
});

route it to Telegram

Message a person, group, or channel through a connected bot.

01
connect a bot
Create a bot with @BotFather and paste its token. We register the webhook and verify it in-region.
02
start a chat
Send /start to the bot from the target chat — or add it to the group/channel — then pick the chat from the list.
03
format the text
Messages use MarkdownV2; the default template bolds the event name and lists fields. Reserved characters in field values are escaped for you.

in telegramdelivered

oncall
*support.ticket.created*
ticket    T-4821
subject   API returning 500s
tier      enterprise
urgency   critical

notes

Telegram caps a bot at roughly 30 messages per second overall, and one per second to a single chat.
The bot must be added to a group — and promoted to admin for a channel — before it can post.
MarkdownV2 requires escaping characters like _ * [ ] ( ); ingestlayer escapes field values, but custom templates are your responsibility.

questions

What model does the classify step use?: Yours. You bring the model, prompt, and label schema; ingestlayer runs it in flight and returns a typed label the pipeline branches on.
Does every ticket hit the model?: Only if you want it to. classify is per-event and cached by payload hash, so identical tickets reuse one call.
Can the same ticket go to two places?: Yes — fan out to several destinations with different when conditions, e.g. critical to chat and everything to Postgres.

build this pipeline or read the quickstart →