Kitsuvo security: detection, data, signing and agents

How detection runs, what leaves your device, how downloads are signed, what a connected AI agent can reach, and where to report a vulnerability.

How detection works

Kitsuvo checks the pages you open on your own device. The browser looks for traces that AI app builders such as Lovable, Bolt and v0 leave in a page: a builder’s script, attribution or hosting address, and unchanged scaffold defaults. A separate slop check looks for content-farm leftovers, such as a chatbot reply pasted into a page. Optional explanations can use local models you install.

The browser’s search filter reads the result links on Google, Bing and DuckDuckGo. Results hosted on known AI-builder domains, and sites the browser has already scored as likely AI-built, fold into one line you can open. It never visits a result you did not open. The extension checks a page only after you choose Scan this page, looks for builder fingerprints only, and its optional search filter hides only results on known AI-builder hosting.

Detection rules ship inside the app and the extension; they are not fetched from a remote list while you browse. A verdict is an estimate from the evidence on the page, not proof of who made it. The disputes page explains what a verdict judges and how a site owner can ask for a review.

What leaves your device

The privacy policy is the full account. These sentences from it cover the browser, the extension, updates and the optional agent tools:

  • Detection: Local checks do not upload page content.
  • No telemetry: The app has no telemetry, analytics or advertising. The app sends us no usage statistics, crash reports or identifiers.
  • Update checks: Like any download, these requests show your IP address and the time to our hosting provider, Vercel. They carry nothing about your browsing, and no identifier for you or your device. To estimate how many installations are active, we look only at aggregate daily and weekly counts of these update checks, calculated within Vercel; we do not export IP addresses or link a check to you or your device.
  • The web engine: Kitsuvo turns off the engine features that would send the addresses you visit to Microsoft, Google or Apple. A few engine settings stay under your operating system’s control; the app section of the policy lists them.
  • The extension: It does not upload page content, collect telemetry, access cookies, or keep browsing history or page reports.
  • A connected AI client: Results returned to the connected client may include personal data in page text, snapshots, reports and supported screenshots. That client handles these results under its own privacy policy.
  • This website: These measurements apply only to kitsuvo.com; the Kitsuvo apps and extensions do not send analytics.

How downloads and updates are signed

The download links on kitsuvo.com redirect to Kitsuvo’s release storage on Vercel, an address ending in public.blob.vercel-storage.com.

  • Mac. The Mac installers are signed with an Apple Developer ID and notarized by Apple.
  • Windows. The current Windows installer, 0.2.4, is not code-signed. Windows may show an unknown-publisher warning, and where Smart App Control is on, Windows blocks the installer with no override. The install guide explains each prompt.
  • Updates. The app accepts updates only if they are signed with Kitsuvo’s update key. You can turn update checks off in Settings › Advanced › Updates.

What a connected AI agent can reach

Kitsuvo’s MCP tools are optional. An AI client can drive the browser only after you add Kitsuvo to that client’s configuration. By default, agent browsing happens in a visible Kitsuvo window with a Stop control; the --headless option hides the window and uses an isolated profile. These rules describe Kitsuvo for Mac 0.2.39:

  • Profile. kitsuvo mcp uses your normal signed-in browser profile by default, so a connected client can use sites you are already signed in to. Add --separate-profile to give the agent a separate persistent profile with its own sign-ins. Separate profiles need macOS 14 or later.
  • Sign-in pages and passwords. Sign-in pages are never read or scripted, and password fields never take typed text, unless you start Kitsuvo with --allow-sensitive.
  • Sensitive sites. Agents are also refused on six categories of sensitive sites, on every profile, unless you start Kitsuvo with --allow-sensitive: webmail; banking, brokerage and crypto accounts; payments and checkout; password managers; cloud consoles; and GitHub, GitLab and Bitbucket settings. The agent can still open such a page, but it cannot read, script, screenshot or act on it, and it is told to ask you to do that step yourself. The site lists are curated, not exhaustive.
  • Cloud relay. The optional cloud MCP relay is started separately. It requires consent in your local terminal and operates an isolated agent profile.
  • Remote access from your other devices. Remote access through the relay and the private-network HTTP bridge both reach only the separate agent profile; a remote session cannot list or open your normal signed-in profile. The bridge listens only on the host's loopback address, needs its bearer token on every request and accepts a network hostname only when you list it in KITSUVO_MCP_ALLOWED_HOSTS. See how to set it up.

The Windows 0.2.4 installer comes from source that predates the sensitive-site categories and the current MCP connection check; the install guide explains what that build supports. Page text an agent reads can contain instructions written to steer it. If you point an agent at sites you do not trust, start it with --separate-profile so it does not share your signed-in sessions.

How to report a vulnerability

Company support contact: kavy@tradewink.com (Tradewink LLC). Include “Kitsuvo security” in the subject.

Email security@kitsuvo.com. The same address is listed in /.well-known/security.txt.

Include the product and version (the Mac or Windows browser, the extension and the browser it runs in, the MCP plugin, or kitsuvo.com), the steps that reproduce the problem, and what an attacker could do with it. Do not send other people’s personal data, real passwords or live API keys; a minimal test page or a redacted capture is enough. Please give us time to fix an issue before you publish its details.

Kitsuvo does not run a paid bug bounty. For a bug that is not a security issue, use Report a bug. For a question about your personal data, use the privacy contact details.