Pipeline

Each episode is a Job. The pipeline runs async on the server; Web and MCP share it.

Stages

1. Ingest          Source plugin or built-in import → SourceArtifact
2. Extract audio   Video demux / audio normalize (text may skip)
3. ASR             Speech → text (text may skip)
4. Spoken script   Transcript/body → structured script + notes
5. Cover           Optional AI image
6. Flashcards      Review cards from content
7. TTS             Script → audio
8. Publish         Library (configurable default)

Input differences

Source	Extract	ASR	Script
Video	✅	✅	From transcript
Audio	Normalize	✅	From transcript
Text / plain	Silent placeholder OK	Skip	From body
URL	Depends on Source kind	By kind	Same

Re-run & reuse

Re-run from a step (e.g. script only or TTS only). Upstream artifacts can be reused. MCP: retry_job.

Failures

• Errors surface on job detail / list
• Transient provider errors → retry
• Config issues → fix Configuration first

Schedules

Schedules only produce candidate URLs and create Jobs — then this pipeline runs.

Code (developers)

• apps/server/src/services/job/
• Media: services/media/ · Content: services/content/

Related

• First episode
• Features
• Architecture