2.1 Data from agent-controlled windows
When (and only when) an agent is executing a task, the Extension collects from the dedicated controlled tab and transmits to your Donkit platform environment:
- Screenshots of the controlled tab (JPEG images of the rendered page);
- Page text and HTML content of the controlled tab, including content inside frames embedded in that page;
- Accessibility-tree data (a structured description of page elements — buttons, links, form fields — used so the agent can interact with the page);
- Page URL and page title of the controlled tab;
- Results of interactions (e.g., which element was clicked, whether an action succeeded or failed, and a short report on the currently focused form field so the agent can confirm that its typing arrived).
This data is sent over an encrypted connection (WSS/HTTPS) to the Donkit platform environment you paired with, where it is used by your AI agent to complete the task you assigned. Processing of this data on the platform — including processing by the large-language-model providers that power your agent — is governed by the general Donkit.AI Privacy Policy.
Sensitive form fields are masked before anything leaves your browser. Where a field is a password field, or is marked by the website as a credit-card field (autocomplete="cc-…") or a one-time code (autocomplete="one-time-code"), the Extension reports only the length of what is in the field, never its contents. This applies both to the accessibility tree and to the focused-field reports, so passwords, card numbers and SMS/authenticator codes are never transmitted to the platform or placed in your agent’s context. Values of other form fields are reported and are truncated to the first 500 characters.
Important: apart from the masked categories above, if you log in to websites or display personal information inside an agent-controlled window, that content may appear in screenshots and page text transmitted to the platform. Controlled windows are separate from your own and are marked while the agent is active, so you can always tell them apart from your personal browsing.
2.2 Data from your normal browsing — not collected
The Extension does not collect, read, record, or transmit:
- your browsing history;
- the content, screenshots, cookies, or credentials of any tab outside the dedicated agent-controlled windows;
- keystrokes or form input in your normal tabs;
- bookmarks, downloads, or other browser profile data.
There is one narrow exception, and it is entirely under your control: when you open the Extension popup to pair your browser, the Extension reads the address of the tab you are currently on and offers it as a suggestion for the Donkit server address, shown as a button you may click. If you do not click it, the address is not used, and it is never transmitted anywhere.
2.3 Pairing and connection data
To establish and maintain the connection, the Extension processes:
- a single-use pairing code that you generate in the Donkit platform and type into the Extension popup; it is consumed by the first successful pairing;
- a device token issued by the platform once pairing succeeds — the credential that lets this browser reconnect without re-pairing;
- a device identifier and a device name. The device name is proposed by the Extension from your browser and operating system (for example, “Chrome on Windows”) and can be edited by you before pairing; it exists so you can recognise this browser in the Donkit interface;
- the address (URL) of the Donkit platform environment you paired with;
- the version number of the Extension, reported when the connection is established, so the platform can tell you when a newer version is available.
These values are stored locally in your browser’s extension storage (chrome.storage.local). They identify the paired browser; the Extension itself does not store your name, email, or Donkit account password.
2.4 Data the Extension does not collect at all
The Extension contains no analytics, telemetry, advertising, or tracking code. It does not use cookies. It does not collect health, financial, payment, location, or biometric information as a category of its own (though, as noted in 2.1, content you choose to display in a controlled window is captured as part of screenshots and page text, save for the masked field categories).