Notebooklm Enterprise

NotebookLM Enterprise API

Trit: 0 (ERGODIC — coordinates knowledge between sources and consumers)

Accessed through the Discovery Engine API (discoveryengine.googleapis.com), not a standalone API.

Interaction with nblm-flashcards skill

This skill is the provider of notebooks consumed by nblm-flashcards:

notebooklm-enterprise (this skill)     nblm-flashcards
────────────────────────────────────    ─────────────────────
Create notebook                    →   Read notebook sources
Add DeepWiki URLs as sources       →   Generate flashcards via Gemini
Check ingestion status             →   Batch per source (5/call)
Share with users                   →   Merge decks across notebooks
Audio overview generation          →   Emacs drill (M-x nblm-drill-all)

GF(3) Triad

notebooklm-enterprise (0) + nblm-flashcards (+1) + drill-verification (-1) = 0

Base URL

https://global-discoveryengine.googleapis.com/v1alpha

All endpoints use project number (not project ID):

projects/302712368086/locations/global/notebooks/{NOTEBOOK_ID}

Active Notebooks (plurigrid/bmorphism ecosystem)

API Quick Reference

Create Notebook

curl -X POST "${BASE}/notebooks" \
  -H "Authorization: Bearer ${TOKEN}" \
  -d '{"title": "My Notebook"}'

Add Sources (batch of 5)

curl -X POST "${BASE}/notebooks/${NB}/sources:batchCreate" \
  -H "Authorization: Bearer ${TOKEN}" \
  -d '{"userContents": [{"webContent": {"url": "https://deepwiki.com/org/repo"}}]}'

Critical: field is webContent.url, not uri, not googleDriveSource.

Get Notebook + Sources

curl -s "${BASE}/notebooks/${NB}" -H "Authorization: Bearer ${TOKEN}"

Share

curl -X POST "${BASE}/notebooks/${NB}:share" \
  -H "Authorization: Bearer ${TOKEN}" \
  -d '{"accountAndRoles": [{"email": "user@example.com", "role": "PROJECT_ROLE_WRITER"}]}'

Audio Overview

curl -X POST "${BASE}/notebooks/${NB}/audioOverviews" \
  -H "Authorization: Bearer ${TOKEN}" -d '{}'

Source URL Rules

Works: DeepWiki URLs, direct PDFs, PMC, arXiv, Frontiers, PLOS, Nature Fails: DOI redirects (doi.org/...), ScienceDirect, paywalled publishers

For the flashcard pipeline, all sources are DeepWiki URLs: https://deepwiki.com/{org}/{repo}

Babashka Pipeline Integration

The backfill pipeline (n/nblm_backfill.bb) automates this skill's operations:

bb n/nblm_backfill.bb --dry-run    # plan: chunk 944 repos into 4 notebooks
bb n/nblm_backfill.bb              # create notebooks + add sources
bb n/nblm_backfill.bb --status     # check ingestion across all notebooks
bb n/nblm_backfill.bb --generate   # trigger nblm-flashcards for each notebook

State persisted at ~/worlds/n/flashcards/backfill-state.json.

GCP Context

Gotchas

1. Use project number not ID in API paths
2. v1alpha only — no v1 or v1beta
3. No list-sources endpoint — sources in GET notebook response
4. Batch size 5-10 for source adds
5. Ingestion is async — poll for SOURCE_STATUS_COMPLETE
6. DOI URLs always fail (no redirect following)
7. Duplicate sources silently accepted
8. Audio overview: empty {} body only, field names rejected
9. Audio has no download API — listen in notebook UI only
10. Podcast API returns 404 despite docs

Related Skills

• nblm-flashcards — Flashcard generation + Emacs drill (PLUS, consumer)
• deepwiki-mcp — Source documentation URLs (ERGODIC, source provider)
• babashka — Pipeline orchestration
• gh-cli — Repo discovery for gap analysis
• vertex-ai — Gemini model access for flashcard generation