nlm-custom-prompt-viewer

Recover and reuse custom prompts used by NotebookLM Studio artifacts from the command line.

[!IMPORTANT] This project depends on notebooklm-sdk (unofficial reverse-engineered API behavior).
Breakages can happen when NotebookLM internals change.

Features

• Extract custom prompts from Studio artifacts:
  • slides, video, audio, quiz, flashcards, report, mind_map, infographic, data_table
• Two extraction modes:
  • Fast (default): direct metadata extraction (method: direct, confidence: high)
  • Infer (--infer): Q&A fallback when direct extraction is unavailable (method: qa_fallback, confidence: inferred)
• Save outputs as json and/or md
• Download artifact outputs with notebook-scoped save paths
• Multilingual CLI messages:
  • default en
  • switchable to ja
• Auth lifecycle commands with secure local storage strategy:
  • keychain-first (keytar, when available)
  • strict file fallback (~/.notebooklm/session.json, 0600 check)
• Cached update notifications (24h window)

Installation

bun install
bun run build

Global usage:

bun add -g nlm-custom-prompt-viewer

Quick Start

1. Authenticate once:

nlmv auth login

2. List prompts in a notebook:

nlmv prompt list <notebookId>

3. Get one artifact prompt and save both formats:

nlmv prompt get <notebookId> <artifactId> --save

4. Download one artifact or all exportable artifacts:

nlmv prompt download <notebookId> <artifactId>
nlmv prompt download <notebookId> a1,a2
nlmv prompt download-all <notebookId>

CLI Overview

Prompt commands

nlmv prompt list <notebookId> [--type <artifactType>] [--json] [--limit <n>] [--infer]
nlmv prompt get <notebookId> <artifactId> [--json] [--save] [--format json|md] [--out <path>]
nlmv prompt download <notebookId> <artifactId|artifactId,artifactId> [--out <path>] [--slide-format pdf|pptx]
nlmv prompt download-all <notebookId> [--out <path>] [--slide-format pdf|pptx]

Config commands

nlmv config get [key]
nlmv config set language [en|ja]
nlmv config reset

Auth commands

nlmv auth status
nlmv auth login
nlmv auth logout

Update commands

nlmv update check

Extraction Modes

| Mode | Flag | method | confidence | Speed | |----------------|-----------|---------------|-------------|-------| | Fast (default) | — | direct | high | Fast | | Infer | --infer | qa_fallback | inferred | Slow |

prompt list defaults to fast mode. Pass --infer to fall back to Notebook Q&A for artifacts where direct extraction yields no result.

prompt get always attempts direct extraction first, then falls back to Q&A automatically — --infer is not required for single-artifact retrieval.

Output Behavior

• prompt list:
  • table output by default
  • machine-readable with --json
• prompt get --save default paths:
  • ./outputs/<slug>.json
  • ./outputs/<slug>.md
  • <slug> = slugify(artifactTitle) (lowercase + hyphens); falls back to artifactId
  • Example: "My AI Podcast" → ./outputs/my-ai-podcast.json
• prompt download / prompt download-all default paths:
  • ./outputs/downloads/<notebookId>/<slug>.<ext>
  • <slug> = slugify(artifactTitle) (lowercase + hyphens); falls back to artifactId
  • slides default to .pdf; use --slide-format pptx for .pptx
  • Extensions:
    • video → .mp4
    • audio → .mp3
    • slides → .pdf or .pptx
    • infographic → .png
    • report → .md
    • quiz / flashcards → .html
    • data_table → .csv
    • mind_map → .json
• --out <path> behavior:
  • Directory (no .json/.md extension): saves <path>/<slug>.<format> for each format
  • File (.json or .md extension) + --format <fmt>: saves directly to <path>
  • Tilde expansion is supported (~/docs → $HOME/docs)
• download --out <path> behavior:
  • prompt download: file extension included = write directly to that file, otherwise treat as directory
  • prompt download-all: always treats --out as a directory
• --format json|md limits output to one format (default: both)
• download progress:
  • binary artifact downloads use cli-progress
  • when Content-Length is available, progress is byte-accurate
  • when it is missing, the CLI falls back to bytes-received display without fake percentages
• Config file: ~/.config/nlm-prompt/config.json (override: $XDG_CONFIG_HOME/nlm-prompt/config.json)
• Session file: ~/.notebooklm/session.json

Interactive Mode

Run without arguments to enter the interactive menu:

nlmv

Arrow-key navigation covers all commands. Required arguments can also be omitted to trigger inline prompts:

nlmv prompt list    # prompts for notebook ID interactively
nlmv prompt get     # prompts for notebook + artifact selection, then offers save/download
nlmv prompt download     # prompts for notebook + artifact multi-selection
nlmv prompt download-all # prompts for notebook selection
nlmv config set     # prompts for key and value

Language Configuration

Default language is en.

nlmv config set language ja
nlmv config set language en

If you omit the value, interactive selection is shown.

Development

With Bun scripts

bun run typecheck
bun run test
bun run build
bun run ci

With justfile

just install
just typecheck
just test
just build
just ci
just prompt-download <notebookId> <artifactId>
just prompt-download-all <notebookId>
just cli "prompt download <notebookId> <artifactId> --slide-format pptx"

Notes

[!TIP] For automation, prefer --json output and pipe results into your own tooling.

[!NOTE] If keytar is unavailable in your runtime, the CLI automatically falls back to secure file-based session storage.

[!NOTE] For tests or debugging, you can force file-only session handling with NLM_PROMPT_DISABLE_KEYTAR=1. This does not change the meaning of NLM_PROMPT_SESSION_FILE; it only bypasses keychain lookup.

[!NOTE] nlm-prompt is an alias for nlmv — both point to the same binary.