Privacy Policy

EdgeAI — Private AI Assistant

No Data Transmission

EdgeAI processes all AI inference (chat, search, embeddings, voice transcription) entirely on your device using WebGPU and WebAssembly. No data is sent to any external server, API, or cloud service during normal operation.

No Accounts or Telemetry

• No user accounts or sign-up required
• No analytics, telemetry, or crash reporting
• No cookies or tracking pixels
• No fingerprinting or device identification

Model Downloads

The only network requests EdgeAI makes are to download AI model weight files from Hugging Face (huggingface.co and its hf.co download mirrors) on first use. The WebGPU model runtime library is bundled inside the extension package; no code is fetched at runtime. Model files are cached locally via the browser's Cache API and are not downloaded again unless you delete them (Settings → Delete Downloaded Models). The network log in EdgeAI's Privacy tab lists every request EdgeAI's own pages make, and you can check that log with Chrome's own tools (see below).

Check the Network Log Yourself

EdgeAI keeps its network log itself: each of its pages records the requests it makes, using the browser's resource timing, and a request appears in the log once it finishes. To check the log with a tool EdgeAI doesn't control:

1. Open chrome://extensions and turn on Developer mode (top right).
2. On the EdgeAI card, under Inspect views, click the link ending in offscreen.html. That is the background page that downloads and runs the models. If it isn't listed, open EdgeAI once and check again.
3. In the DevTools window that opens, choose the Network tab and use EdgeAI as usual. Every request that page makes appears there, and should match the log.

The popup works the same way (right-click inside it, then Inspect), as does the service worker (its link on the same card). DevTools records only while it is open.

Data Stored Locally

All user data is stored in your browser's local storage (IndexedDB and chrome.storage.local):

• Documents and chunks: Text content you import (Obsidian vaults, PDFs, bookmarks, web pages) is chunked and stored with vector embeddings for retrieval
• Chat history: Your conversations with the AI assistant
• Settings: Your voice and display preferences
• Audit log: A local record of all queries and operations (viewable in the Trust Panel)

Clearing Your Data

You can delete EdgeAI data at any time:

• Settings → Clear All Data removes every document, chunk, embedding, chat session, log, setting and saved folder permission from this device
• Settings → Delete Downloaded Models removes the cached model files (about 2.4 GB)
• Remove individual documents from the Documents tab, or individual chats from the history panel
• Clear the audit log and network log from the Trust Panel
• Uninstalling the extension removes everything, including model files

Permissions Explained

EdgeAI has no access to the websites you visit. It reads a page only when you ask it to index that page.

• storage / unlimitedStorage: Store your documents, embeddings, chats and settings on this device, and cache the model files
• offscreen: Run the AI models in a background document (required for WebGPU access)
• activeTab / scripting: Read the text of a page when you index it. Chrome allows this only on the tab where you opened EdgeAI (toolbar icon, keyboard shortcut or right-click menu), and only until that tab goes to another page
• contextMenus: Add "Index this page with EdgeAI" to the right-click menu
• notifications: Tell you when a page you right-clicked is being indexed or couldn't be read, and when a bookmark import finishes
• sidePanel: Open EdgeAI in Chrome's side panel
• bookmarks (optional): Asked for only when you choose Import → Chrome Bookmarks. EdgeAI imports each bookmark's title and address; it doesn't visit the pages

Model files download from Hugging Face without any website permission: Hugging Face allows downloads from other origins, so none is needed.

Open Source

EdgeAI is fully open source. You can inspect every line of code, verify the build, and confirm that no data exfiltration occurs. The exact commit hash of the running build is shown in the Trust Panel.

Contact

Questions about privacy? Open an issue on GitHub. Source code: github.com/kamalkalwa/EdgeAI.