RAG pipeline
Turn explicit video lists into JSONL-ready source bundles.
For RAG and evaluation work, the useful artifact is not a chat answer. It is a reproducible bundle with per-video JSON, optional JSONL, combined Markdown, and a manifest that records status and billing facts.
Schema-first output
Use manifest.json as the source of truth and videos.jsonl as an ingestion-friendly export.
Partial failure visibility
Each item records status, output paths, and billing facts so failed items can be retried intentionally.
No hidden crawling
The first beta processes the URLs you pass. Discovery, channel watch, and hosted batch jobs are later design topics.
Workflow
How the beta flow works
- Collect the public video URLs you are allowed to process.
- Run the CLI with combined Markdown and JSONL exports enabled.
- Load videos.jsonl into your indexer and keep manifest.json for audit and retries.
Sample bundle
Inspect the output shape before requesting access.
These are synthetic demo files, not customer records. They show the stable bundle contract: manifest first, then per-video and aggregate outputs.
Current boundaries
What this page is not promising.
- No hosted remote MCP endpoint is launched by these pages; the current agent path is local stdio.
- No server-side batch dashboard or queue is promised here.
- No automatic playlist expansion, channel crawler, or YouTube search is included in the first beta.
- Paid API usage and billing decisions remain server-authoritative.