Skip to main content

How ingest works (and why it takes hours)

Every episode in your feed goes through three stages before it’s available to answer questions:

  1. Transcribe — audio → text. YouTube captions are used when available (fast). Everything else is transcribed via Whisper (slower).
  2. Embed — text → vector representation, chunked to ~30–60 second utterances. Enables semantic search.
  3. Enrich — structured extraction (speakers, series, bible verses for sermons, topics) via Gemini.

Typical wait

Feed shapeTime to “ready”
YouTube channel with clean captions, ~20 videosUnder 1 hour
RSS podcast, 20 episodes, audio only (Whisper)3–6 hours
Podcast with 200-episode backlog1–3 days full backfill; ready state hits at ~5 episodes

What “ready” means for your trial

Your 14-day paid trial clock starts when your feed is ready to answer questions well — not at signup. See Trial and plans for the details.

We define “ready” as: 5 episodes have completed all three stages (transcribe + embed + enrich). At that point the widget can produce reasonable answers.

Watching progress

The Ingest Progress panel on your /console dashboard polls live status. You’ll see episode counts, per-stage state, and which episode is currently processing.

This is a stub page — expanded content coming in a follow-up. Cross-references from other docs point here.

Related