teng-lin/notebooklm-py v0.8.1

on GitHub

Added

• Added account-level notebook collections through client.collections and the
  notebooklm collection CLI group, including list, create, rename, membership,
  and delete operations.
• Expanded source discovery with status/type filters, strict exact counts, Drive
  health, original-file links, MIME information, word counts, and revision
  timestamps. MCP and REST URL batches now use a single batch-capable request.
• Added structured source documents and offset-aware citation data so callers can
  render readable source text and align citations with both answers and sources.
• Exposed more notebook, artifact, chat, and research metadata, including project
  state, chat follow-up suggestions, artifact content/state, research mode,
  timings, termination reasons, and discovered-source ordinals.
• Added standalone notebooklm research import, configurable research-import
  timeouts, bulk sharing grants with SharingAPI.set_users(), and sharing-policy
  and collaborator-limit fields.
• Added opt-in mid-session NOTEBOOKLM_REFRESH_CMD support for long-running
  servers with NOTEBOOKLM_REFRESH_CMD_MIDSESSION=1.

Changed

• The default NotebookLM host is now https://notebook.google.com; the previous
  https://notebooklm.google.com host remains supported for existing setups.
• Notebook.last_viewed_at and NotebookMetadata.last_viewed_at now describe the
  backend timestamp accurately. The old modified_at names remain compatible
  aliases through v0.x.
• Deep-research CLI waits now default to 1,800 seconds instead of 300 seconds.
• Compatibility note: QuizQuantity.MORE now has its own wire value and no
  longer compares equal to QuizQuantity.STANDARD. Quiz and flashcard option
  inputs are also validated as enums.

Deprecated

• Direct AuthTokens storage loading and cookie projections are deprecated in
  favor of managed NotebookLMClient.from_storage() lifecycles. They remain
  available through v0.x and are scheduled for removal in v1.0.
• The pre-profiles home-root storage layout is deprecated. Running any
  notebooklm command migrates it to profiles/default/ automatically.
• ChatReference.answer_start_char / answer_end_char are deprecated aliases
  for fragment_start_char / fragment_end_char; the old names remain through
  v0.x.

Fixed

• Improved authentication and login reliability across both NotebookLM hosts,
  including Chromium setup diagnostics, custom storage paths, expired-session
  recovery, cookie persistence, and long-running MCP/REST sessions.
• Prevented notebook and source create retries from adopting an older item,
  duplicating a committed item, or retrying when the idempotency probe itself
  could not establish a safe answer.
• File uploads now preserve partial-failure recovery details, report processing
  failures that never resolve, and warn when a mistyped file extension would be
  treated as pasted text.
• Corrected chat answer selection, conversation turn numbering, Unicode and
  multi-block citation offsets, full cited-text extraction, and source passage
  resolution.
• Corrected notebook ownership, not-found handling, empty-notebook decoding, and
  create results when the pre-create baseline is unavailable.
• Corrected artifact status decoding, generation rejection messages, quiz and
  flashcard option ordering, and audio downloads (.m4a / audio/mp4 rather
  than .mp3).
• Research imports now tolerate the backend's known retry response, distinguish
  no-results and cancellation outcomes, and preserve typed failure reasons.
• Corrected PowerPoint and unknown source-status decoding, Drive-source
  idempotency, Markdown full-text formatting, table-cell rendering, and custom
  per-RPC timeout handling.

Install with pip install notebooklm-py==0.8.1. Full changelog: CHANGELOG.md.