Last updated: 2026-06-27

Notch is a local-first, bring-your-own-key (BYOK) browser extension. It is built so that your data stays on your device and is only ever sent to services you explicitly configure. Notch has no backend server, performs no analytics or telemetry, and never sends your data to the developers.


All of the following is stored locally on your device (browser IndexedDB and browser.storage.local). None of it is transmitted to Notch or any third party except as described under "Network connections" below.

| Data | Purpose | Location |
| --------------------------------------------------- | --------------------------------------------------- | ------------------------------------------------- |
| Captured page content, summaries, notes, highlights | The knowledge base you build | IndexedDB (notch_db) |
| Embeddings / search index | Semantic & full-text search | IndexedDB |
| Provider API keys | Authenticating to the AI provider you configure | IndexedDB / browser.storage.local (unencrypted) |
| Settings & appearance | Your preferences | browser.storage.local |

API keys are stored unencrypted in the browser's local database. Anyone with access to your browser profile can read them. Clear them at any time in Settings → Providers → Clear all API keys, or remove an individual provider.


Notch only makes network requests in these cases, all driven by your actions or configuration:

1. Your AI provider — When you capture or chat (and are not in local-only/on-device mode), the relevant page content and your prompt are sent to the AI provider endpoint you configured (e.g. OpenAI, Anthropic, Google Gemini, OpenRouter, Groq, or any OpenAI-compatible/Ollama endpoint), authenticated with your key. Notch does not proxy or see this traffic.
2. On-device model downloads (opt-in) — If you enable On-Device AI and choose a model, the model weights are downloaded on demand from the Hugging Face Hub (huggingface.co). Models are not bundled in the extension. Nothing is sent — only downloaded.
3. Capturing arXiv / Semantic Scholar pages — When you capture one of these pages, Notch fetches the corresponding PDF from arxiv.org / semanticscholar.org to extract its text.
4. Media thumbnails — Capturing a page with embedded YouTube content may load thumbnails from img.youtube.com.
5. PDF export styling — Exporting a document to PDF loads the KaTeX math stylesheet from cdn.jsdelivr.net into the print document.
6. Remote PlantUML diagrams (opt-in, off by default) — Only if you explicitly enable it in Settings, PlantUML diagram source is sent to plantuml.com to render an image. Disabled by default; always suppressed when the Local-only lock is on.

When the Local-only lock is enabled in Settings, Notch performs no network calls at all — capture and chat use the bundled offline NLP and any installed on-device models.

• ❌ No analytics, telemetry, tracking, or "phone-home" of any kind.
• ❌ No advertising or selling of data.
• ❌ No remote code execution — all executable code is packaged in the extension.
• ❌ No transmission of your content or keys to the developers or any server we operate.

• Delete any captured document from the Library at any time.
• Clear all stored API keys in Settings → Providers.
• Uninstalling the extension removes all locally stored data.

• storage — persist your knowledge base and settings locally.
• tabs / activeTab — read the current page when you ask Notch to capture it.
• scripting (Chrome) — inject the extraction logic into the page you are capturing.
• contextMenus (Chrome) — the right-click "Capture with Notch" entry.
• host_permissions (*://*/*) — capture works on any site you choose; the content script only runs to extract a page when you invoke capture.
• debugger + downloads (Chrome, optional) — requested only when you use one-click PDF export, never at install time.

For privacy questions, open an issue on the project repository at https://github.com/CR-8/Notch.