NotebookLM has spent the past 12 months climbing Google Trends with a trajectory no other AI tool has matched. Not a viral spike – a sustained, month-over-month growth curve that reflects genuine, recurring user demand. Researchers use it to interrogate their own documents. Students use it to generate study guides from lecture recordings. Knowledge workers use it to turn 200-page reports into 10-minute podcast briefings.

And yet: there is no desktop app.

Google ships NotebookLM as a web app at notebooklm.google.com and as an iOS app on the App Store. On macOS, Windows, and Linux, the only option is a browser tab. GeminiDesktop exists to close that gap – a native desktop app (Windows, Mac Intel + Apple Silicon, Linux) that brings the full NotebookLM experience to your desktop, built on Tauri and the official Gemini API.

TL;DR / Key takeaways

NotebookLM has no native desktop app on any OS as of April 2026 – only a web UI and an iOS app. No Mac app, no Windows app, no Linux binary, no Android app.
The Gemini for Mac app (2026-04-15) advertises NotebookLM integration on its landing page, but the accompanying blog post and third-party coverage stop short of confirming full feature parity. Status remains ambiguous.
GeminiDesktop.app is a native Tauri 2.x client that ships NotebookLM-style document workspaces natively: right-click a folder, open it as a notebook, generate Audio Overviews saved as MP3 to ~/Music/, export Mind Maps to Obsidian Canvas, render Video Overviews locally.
The core design shift is “a folder is a notebook” – no uploads, no 20 MB cap, automatic re-index when files change. Works with Obsidian, Logseq, Zotero, Typora, or any folder-based workflow.
Windows users get parity for the first time – Google has no native Gemini Windows app either. GeminiDesktop is the only real native Windows option today.
Why hasn’t Google shipped this? Structural incentives inside Google Labs keep NotebookLM as a small-team, web-only product with no public API. We dig into this below.

Why NotebookLM needs a desktop app

The browser is adequate for one-off queries. It is not adequate for the way power users actually work with NotebookLM.

A researcher preparing a literature review has 40 PDFs in a local folder. In the browser, each one must be uploaded individually through a web form, and anything over 20 MB is rejected. On the desktop, the folder itself becomes the notebook – drop your files in, and they are indexed automatically with no size ceiling.

A podcast producer generates Audio Overviews from client briefs every morning. In the browser, the MP3 plays in-page with no download button that writes to the local filesystem. On the desktop, the file saves directly to a configured folder where Apple Music, Spotify local files, or any media player can pick it up.

A student working with Obsidian or Logseq has hundreds of markdown notes organized in a vault. In the browser, those notes must be copy-pasted or uploaded one by one. On the desktop, the vault folder is the notebook. Every markdown file is a source. When you add a new note, it appears in the index within seconds.

A developer onboarding to a new codebase points the app at a documentation folder and a few architecture decision records. On the desktop, the notebook updates as git pulls land new files. In the browser, they’d upload the same bundle twice a week.

A legal team reviewing discovery documents – PDFs, emails, scanned contracts – wants everything indexed locally, not uploaded to a Google product subject to Google’s retention policies. The browser version makes that impossible. A native desktop client keeps the corpus on your machine and sends only queries to the API.

These are not hypothetical scenarios. They are the workflows that NotebookLM’s own subreddit discusses every week, and the workflows that the browser version cannot support.

12 months of sustained growth, zero desktop investment

Pull up Google Trends and compare “NotebookLM” against any other AI-related desktop term. “Gemini desktop.” “Claude desktop.” “ChatGPT desktop.” NotebookLM’s search interest runs two to three times higher than all of them combined, and the curve is still climbing as of April 2026.

This is not a product that went viral for a week and faded. This is a product with compounding organic demand. The kind of demand that, in any normal product organization, would trigger immediate platform expansion.

Google’s response: an iOS app (real, functional, in the App Store) and nothing else. No Mac app. No Windows app. No consumer API. The Gemini for Mac app that launched on April 15, 2026, lists “NotebookLM” on its landing page, but no reviewer has demonstrated full NotebookLM functionality – Audio Overview, Video Overview, Mind Map, Interactive Mode – working through it. And the Google App for desktop that landed on April 14, 2026 for Windows ships without any NotebookLM surface at all.

Why Google hasn’t shipped it – the strategic read

NotebookLM lives in Google Labs with a tiny team that punches far above its weight. A few structural dynamics explain why the desktop app hasn’t landed:

No public API. Without an API, there is no external ecosystem pushing Google to build a reference desktop client. Every other successful AI product (OpenAI, Anthropic, Perplexity) has a public API that third parties use to build desktop integrations. NotebookLM has only its own web UI. The lack of an API is a choice – and it limits internal pressure to ship shells around it.
Labs products are experiments first. Inside Google, Labs products are judged on learnings, not platform coverage. A Mac app requires macOS signing, notarization, auto-update pipelines, crash telemetry, and customer support. None of those are core to a Labs learning goal.
The “one team, one codebase” web-first culture. Shipping native desktop apps requires OS-specific expertise that a web team usually doesn’t have on staff. Spinning up a Mac/Windows/Linux trio means hiring or borrowing teams – another structural speedbump.
Business incentives unclear. NotebookLM Plus monetizes through Google AI Pro. Pro users are already paying for Gemini + Workspace. A desktop app doesn’t unlock new revenue streams that obviously justify the engineering cost.

We cover the full structural analysis in Why NotebookLM Has No Native App and The NotebookLM Mac app Google still hasn’t made.

One folder, one notebook

The core design principle of GeminiDesktop’s NotebookLM integration is simple: a local folder is a notebook.

You point GeminiDesktop at ~/Research/climate-papers/. Inside that folder are PDFs, markdown notes, audio recordings, images, and a few YouTube URLs saved as text files. The app indexes everything. Each file becomes a source. The chat interface is grounded in that content – when it answers your questions, it cites the specific documents and passages that support the answer.

This paradigm has three properties that the browser version cannot match:

No upload friction. Files are read from disk. There is no upload step, no file-size limit imposed by a web form, no waiting for a progress bar. A folder with 100 PDFs indexes in the time it takes to scan the filesystem and extract text.

Live sync. When you save a new file into the folder, it appears in the notebook. When you delete a file, it disappears from the index. The notebook reflects the current state of the folder at all times. There is no “re-upload” step.

Compatibility with existing tools. If you use Obsidian, your vault is already a folder. If you use Logseq, your graph is already a folder. If you use Typora, your documents are already in a folder. GeminiDesktop does not ask you to migrate your files into a proprietary format or a cloud service. It reads what you already have, where you already have it.

Audio Overview, Video Overview, and Mind Map – saved locally

NotebookLM’s Studio panel generates three types of rich output: Audio Overviews (podcast-style conversations), Video Overviews (narrated visual summaries, now assisted by Veo-3 generation), and Mind Maps (concept relationship graphs).

In the browser, these outputs exist inside the NotebookLM interface. You can play an Audio Overview in the browser tab. You can watch a Video Overview in the browser tab. You can explore a Mind Map in the browser tab. But you cannot easily extract them for use elsewhere.

GeminiDesktop saves these outputs to your local filesystem:

Audio Overview saves as an MP3 to ~/Music/GeminiDesktop/ or any folder you configure. macOS indexes the file. Apple Music sees it. Your iPhone syncs it via iCloud. CarPlay plays it on your commute. On Windows, the MP3 drops into Music\GeminiDesktop\ and gets indexed by Windows Media Player or whatever audio app you prefer.
Video Overview saves as an MP4 to ~/Movies/GeminiDesktop/ or your configured path. Share it with colleagues, upload it to a CMS, or import it into a video editor.
Mind Map exports as SVG, PNG, interactive HTML, or Obsidian Canvas .canvas JSON. Embed it in a research paper, paste it into Notion, or open it in Obsidian for full interactivity.

The generated content belongs to you. It lives on your machine. It does not require an internet connection to access after generation.

Use case walkthroughs

Academic researcher – literature review in days not weeks

Drop a folder of 40 PDFs into ~/Research/climate-adaptation/. Open the folder as a notebook. Ask “What’s the current consensus on soil carbon sequestration effectiveness?” with cited answers referencing specific papers. Generate a Mind Map to see how the papers relate. Export to Obsidian Canvas and link each node to your own running notes. Generate a 15-minute Audio Overview in “debate” style to stress-test the consensus. The whole workflow stays on disk.

Journalist – briefing prep for a 2pm call

You have a 60-page regulatory filing you need to understand before a source call at 2pm. Open it with GeminiDesktop. Ask “What are the three things in this filing that didn’t exist in last year’s version?” with citations back to the specific pages. Generate a 5-minute Audio Overview to listen to on the walk from your apartment to the coffee shop. Bring the cited answers into your notes app of choice.

Developer onboarding – shortcut the “read the codebase” stage

Point the app at your new team’s docs/ directory and a curated set of key source files. Ask architecture questions and get grounded answers with file:line citations. Generate a Mind Map of the key subsystems. Within an afternoon you have a structural view of the system that would normally take a week of reading.

Competitive analysis – synthesize 20 competitor pages

Save 20 competitor landing pages, press releases, and pricing pages as PDFs into a folder. Ask “Who positions themselves against us, and what claims do they make?” Get cited answers. Generate a Mind Map of competitor positioning clusters. Export it to share in Notion. Re-run monthly as the folder auto-indexes new files you drop in.

Legal document review – local-only discovery

For privacy-sensitive discovery, configure the folder inside an encrypted volume. Queries still go to the Gemini API, but the source corpus itself never gets uploaded to a Google product subject to Google’s general retention policies. The native desktop posture gives you control that the browser can’t offer.

How GeminiDesktop implements NotebookLM features

GeminiDesktop is built on Tauri 2.x, a Rust-backed framework that produces native apps under 15 MB. It connects to the Gemini API directly – the same API surface that powers NotebookLM’s web interface – and implements the core NotebookLM workflows locally.

Source ingestion

When you open a folder as a notebook, GeminiDesktop scans the directory recursively. It extracts text from PDFs (via pdf-extract), parses markdown and plain text files, transcribes audio files, and processes images through Gemini’s multimodal input. Each source gets an entry in a local SQLite index with extracted text, metadata, and embeddings for retrieval.

Grounded Q&A

When you ask a question, GeminiDesktop retrieves the most relevant passages from your indexed sources, sends them as context to Gemini 3, and returns an answer with citations pointing back to specific files and page numbers. This is the same retrieval-augmented generation (RAG) approach that makes NotebookLM’s web version reliable – answers are grounded in your actual documents, not in the model’s general training data.

Studio panel outputs

Audio Overview, Video Overview, and Mind Map generation all run through Gemini’s API. The app sends your source content, receives the generated output, and writes it to disk. Audio Overviews use Gemini’s multi-speaker voice synthesis (the gemini-3.1-flash-tts-preview model with MultiSpeakerVoiceConfig) to produce natural two-host conversations. Video Overviews combine Gemini-generated scripts with Veo-3-assisted visuals and render locally via Remotion. Mind Maps are constructed from extracted concept relationships and rendered locally with ReactFlow.

Watch-folder automation

You can configure any folder as a “watched” directory. When new files appear – from a browser download, a Zotero sync, an email attachment save – the app automatically indexes them into the appropriate notebook. Optionally, it can trigger an Audio Overview generation on a schedule (e.g., 3:00 AM local time), so you wake up to a podcast briefing of everything that arrived overnight.

Multi-model, not Gemini-only

GeminiDesktop is not a Gemini-only client. It ships with multi-model support: Gemini 3 for the heavy RAG work, Claude Sonnet 4 for dense technical reading, and GPT for general chat. You configure which model answers which kinds of questions, or you toggle per-session. NotebookLM itself is Gemini-locked. GeminiDesktop is not.

Compatible with your existing tools

GeminiDesktop does not replace your note-taking app. It augments it.

Tool	How it works with GeminiDesktop
Obsidian	Point GeminiDesktop at your vault folder. Every `.md` file becomes a source. Export Mind Maps to `.canvas`.
Logseq	Same approach – your `pages/` and `journals/` directories are indexed.
Typora	Your document folder is the notebook.
Zotero	Configure Zotero’s attachment folder as a watched directory. New papers auto-index.
Apple Notes	Export to markdown (via Exporter), then index the output folder.
OneNote / Evernote	Export to `.pdf` or `.md`, drop into a folder.
Git repos	Point at a docs directory; fresh commits auto-ingest.

The principle is the same in every case: if your tool stores files in a folder, GeminiDesktop can read them. No plugins, no integrations, no API keys to configure. The filesystem is the integration layer.

Windows context – no first-party alternative

If you’re on Windows, the situation is even starker. Google has not released a native Gemini Windows app. The Google App for Windows (2026-04-14) ships without NotebookLM integration. Microsoft’s Copilot is the Windows-native AI client, but it’s OpenAI-backed and doesn’t replicate NotebookLM’s grounded document workspaces. GeminiDesktop’s Windows build gives you real native performance and NotebookLM-style workflows on a platform Google has ignored. See Why Google didn’t make a Gemini Windows app and the Windows install guide for more.

What you get that the browser does not offer

Capability	NotebookLM Web	GeminiDesktop
Local folder as notebook	No	Yes
Auto-index new files	No	Yes
Audio Overview saved as MP3	No	Yes
Video Overview saved as MP4	No	Yes
Mind Map export (SVG/PNG/HTML/Canvas)	No	Yes
Watch-folder automation	No	Yes
Works with Obsidian/Logseq vaults	No	Yes
Offline access to generated content	No	Yes
Native macOS app (Intel + Apple Silicon)	No	Yes
Native Windows binary	No	Yes
Native Linux binary	No	Yes
Interactive Mode via global hotkey	No	Yes
Multi-model (Gemini + Claude + GPT)	No	Yes
20 MB file size cap	Yes	No (no cap)

Limitations

No tool is perfect. GeminiDesktop’s Audio Overview, Video Overview, and Mind Map all depend on the Gemini API, which means API quota costs apply and an internet connection is required for generation (reading cached outputs is offline). Interactive Mode requires a stable connection for Gemini Live’s WebSocket stream. Some file formats (.pages, .key, .numbers) require export-to-PDF before indexing. And while the app is localized for English, Chinese, Japanese, and Korean UIs, not every language feature matches NotebookLM’s 80+ generation languages perfectly yet.

FAQ

Does NotebookLM have a desktop app in 2026? No. As of April 2026, Google has not released a native NotebookLM app for Mac, Windows, or Linux. Only web and iOS.

NotebookLM vs Gemini – what’s the difference? Gemini is the general-purpose chatbot. NotebookLM is a specialized RAG product that grounds answers in documents you upload. Both are Google products, both use Gemini models, but NotebookLM adds source indexing, Audio/Video Overviews, and Mind Maps.

Can I use NotebookLM offline? No. NotebookLM needs an internet connection for everything. GeminiDesktop caches generated outputs (MP3, MP4, SVG) so you can re-read them offline, but new generations still call the API.

Is NotebookLM free? Yes, with limits – 50 sources, 50 audio generations per day. NotebookLM Plus (Google AI Pro) raises limits and unlocks shared notebooks and analytics.

What files can NotebookLM read? PDFs, Google Docs/Slides, URLs, YouTube videos, plain text, Markdown, pasted text. Max 20 MB per file. GeminiDesktop reads the same plus more, without the 20 MB cap.

Is GeminiDesktop official Google software? No. GeminiDesktop is an independent native desktop client that uses the official Gemini API with a BYO-key model. Positioning: “what Google forgot to build.”

Try it

GeminiDesktop is available for macOS (Intel + Apple Silicon), Windows, and Linux. Download it, point it at a folder, and see what NotebookLM feels like when it runs natively on your machine instead of in a browser tab.

geminidesktop.app/app