Zero egress, by construction — not by policy.

01The guarantee, in one sentence

The filing text Relic reads, and every insight it derives from it, never leaves your device. The extension runs no servers, sends no telemetry or analytics, and holds no accounts. The only outbound connection it can make is an HTTPS GET to SEC.gov for a public document — and that is optional and user initiated. Everything below is how that is enforced.

Three things you can check rather than take on faith:

• It runs on your machine. The models are bundled with the extension and run in your browser, so nothing about your filing is sent out to be analyzed.
• It talks only to SEC.gov. Outbound connections are locked at the browser level to SEC.gov — enforced by the extension's own configuration, not a toggle you have to trust.
• It fails loudly if that ever weakens. A build time gate blocks any release that loosens the lock, so the privacy claims reduce to things you can confirm in the Network tab.

02The Content Security Policy lock

Relic pins a Content Security Policy that restricts all network connections to SEC.gov and forbids remote script. This is the load bearing control — the browser itself refuses any connection outside the allowlist, regardless of what the code attempts. There is no analytics domain, no telemetry endpoint, and no cloud AI host it could ever talk to, and no script is fetched or evaluated from any remote source.

03The build time gate that keeps it honest

A guarantee is only as good as the discipline that maintains it. Relic's zero egress properties are enforced automatically at build time — the release fails if any of them (the CSP network lock including default-src 'none', the bundled models and ONNX runtime, host access limited to SEC.gov, and COOP/COEP isolation) weakens. A degraded privacy posture can't be shipped by accident; it's an invariant, not a matter of remembering to check.

04On device by design

All analysis runs locally. Models are bundled into the extension and loaded from inside the package — never fetched at runtime — so inference never requires, and the CSP never permits, a call to a remote model host. Generative summaries use Chrome's own on device AI when the device provides it, with a deterministic fallback otherwise, and every generated statement is checked against the source filing so claims can't drift from what the document actually says.

05The one network path — and its off switch

Within the CSP allowlist, Relic makes exactly one kind of request: an HTTPS GET to SEC.gov to fetch the prior year public filing when you run a year over year Redline. These requests carry no personal data — they are ordinary requests for public documents. You can turn this off in Settings → "Fetch prior year filings from SEC.gov," after which Relic makes no network requests at all. Optional host permissions for a few finance sites let Relic read a page you already opened; they do not expand connect-src. The only extension-initiated HTTPS request remains SEC.gov.

06Scope and honest limits

To keep this page defensible, here is what Relic does not claim:

• Relic secures its own behavior — it does not harden your browser, OS, or the SEC.gov connection beyond using HTTPS. It is not a substitute for endpoint security controls.
• The zero egress guarantee covers the extension as shipped. As with any software, it depends on installing the genuine build from the Chrome Web Store and on Chrome's own extension sandbox.
• The prior year filing fetch, when enabled, reveals to SEC.gov that a public document was requested — the same as opening that document in a browser tab. It carries no identity or analysis data.