REPORT FEED // PRIVACY_ NODE:
LOG OPENED · SOURCE
PRIVACY POLICY
What reads from your pages, where it sends it, and what stays on your device.
BYOK — YOUR AI PROVIDER, YOUR DATA
MEMORY & SETTINGS STORED LOCALLY
DEFAULT DEMO ENDPOINT SEES PAGE DATA
REPORT: *** SCOPE- What does ***
Summary: is a browser automation agent. When you run a task, it reads the active page's DOM, builds a simplified/cleaned text representation, and executes the actions you asked for (clicking, typing, scrolling, navigating), including across multiple tabs. Built-in skills use the same page-reading approach for specific workflows: cleaning up Gmail/Outlook inboxes, organizing Google Calendar, tracking a financial watchlist, summarizing news, scanning resumes, and preparing executive reviews.
  • The task instructions you type or speak.
  • A simplified/cleaned copy of the page structure and visible text of the tab(s) you run a task on.
  • For email/calendar skills: the relevant message, event, or thread content visible on the page at the time.
Addendum: only reads pages while a task is actively running or a skill is explicitly invoked- it does not passively monitor your browsing.
— ✦-
REPORT: *** WHERE YOUR DATA GOES ***
Summary: Task instructions and the cleaned page content described above are sent to whichever AI provider you configure in Settings, so the model can decide what actions to take.
  • OpenAI / Anthropic / other API providers: sent directly from your browser to that provider using the API key you supply. Their privacy policies apply.
  • Your own local model (e.g. Ollama on localhost or your LAN): stays entirely on your network.
  • In-browser model: runs fully on-device; nothing is sent over the network.
  • Default demo endpoint (if you haven't configured a provider): requests go to our hosted testing API, run on Alibaba Cloud infrastructure, for evaluation purposes only. It does not log, store, or train on your prompts or page content- data is processed in-transit and discarded. We recommend connecting your own API key before using the extension on pages with sensitive data. Full terms: terms & privacy.
Addendum: does not operate a backend that stores your task content. We only stand up the optional evaluation endpoint described above.
— ✦-
REPORT: *** PAGE & SITE ACCESS ***
Summary: requests broad host permissions because it needs to act on whatever page you point it at- there's no fixed list of supported sites. Page reading and DOM actions only happen while you are actively running a task or an explicitly-invoked skill.
  • Gmail / Outlook cleanup: reads inbox/message content on those pages only when you run the skill; performed by DOM automation in your own logged-in session, not a separate Gmail/Outlook API integration.
  • Google Calendar: reads and reorganizes events visible on the calendar page when you run the skill.
  • General automation tasks: reads and acts on whichever tab(s) you direct the agent to.
Addendum: The extension does not read your passwords or autofill credentials, and does not access tabs you have not asked it to act on.
— ✦-
REPORT: *** MEMORY & OPTIONAL INTELLIGENCE FEATURES ***
Summary: keeps a local memory of past sessions to give more relevant help over time, and includes optional on-device ML features. Neither sends your data to .
  • Episodic memory, synthesized notes, and session history are stored in your browser's IndexedDB.
  • Nothing in the memory store is synced to a remote server by default.
  • If you connect an MCP vault (e.g. a local Obsidian server), memory can be exported there. This is a destination you configure yourself, typically on localhost.
Addendum: On-device ML (embeddings, clustering) used for memory search runs entirely inside your browser and does not transmit data externally.
— ✦-
REPORT: *** WHAT WE DO NOT DO ***
  • We do not operate a backend that stores your task content, page data, or memory (only the optional evaluation demo endpoint exists, and it discards data in-transit).
  • We do not collect telemetry, analytics, crash reports, or usage statistics.
  • We do not show ads inside the extension.
  • We do not sell, share, or monetize your data in any form.
  • We do not use your page content or task instructions to train any AI model.
  • We do not execute any remote code- all extension logic is bundled locally at build time.
— ✦-
REPORT: *** RETENTION & DELETION ***
Summary: Your configuration, memory, and history live in your browser. holds no copy to delete on your behalf, except transient in-memory processing on the optional demo endpoint, which is discarded immediately after the request.
  • chrome.storage.local: your AI provider settings and API key, feature toggles. Cleared on uninstall.
  • IndexedDB: session history and memory (episodic notes, synthesized context). Cleared on uninstall.
  • OPFS / on-device model cache: local ML model weights and embeddings, if used. Cleared on uninstall.
Addendum: If you configured a third-party AI provider or your own MCP vault, deleting data held there requires using that service's own tools- cannot reach back and delete requests it already sent.
— ✦-
REPORT: *** YOUR CONTROLS ***
  • Provider choice: connect your own OpenAI, Anthropic, local, or in-browser model at any time in Settings- this stops requests to the default demo endpoint immediately.
  • Skill toggles: enable or disable individual skills (email cleanup, calendar, financial monitor, etc.) in the Agent Library.
  • Clear local data: uninstalling the extension removes all locally stored settings, memory, and history from your device.
  • Vault opt-out: the MCP vault integration is off by default and only sends data where you explicitly point it.
Full terms & privacy:
Extension homepage:
— ✦-
REPORT: *** CHANGES & CONTACT ***
If this policy changes in a way that affects how your data is handled, we will show an in-extension notice before the change takes effect and update the effective date at the top of this log.
Questions or requests:
— END OF LOG-