Ingest

First stage of the core pipeline: accept webhook payloads and write graph entries to Neo4j.

Flow


POST /api/story/webhook
  → optional forward (see Resolvers)
  → lpush timelining::ingest::backlog
  → dispatch /api/story/ingest?limit=1&mode=chain
  → Neo4j Entry (+ Voice, Participant, Chat, …)
  → post-ingest actions (transcribe chain or resolve trigger)

Non-reply messages are queued and ingested. Replies may forward to siblings but skip the backlog.

Webhook filter

Messages are processed when the chat is private, the topic name contains _bot, or the topic is prisma_events_storying (messages sent to bot directly).

Ingest worker

/api/story/ingest pops the Redis backlog (services/ingest/):

Maps Telegram payload → FullEntryInputData (lib/db/mappers.ts)
Creates nodes via entryService / lib/db/models/
Runs pipelineActionsAfterIngest — transcribe dispatch or resolve trigger

Modes: chain (one message, inline continuation), batch (cron sweep), retry (failed queue).

Durability

Queue	Purpose
`timelining::ingest::backlog`	Pending messages
`timelining::ingest::failed`	Write failures (retry cron)

Redis is provisioned via propagate (upstash.kv). No separate setup doc needed.

Code

src/app/api/story/webhook/, src/services/ingest/, src/services/pipeline/routing.ts