Limitations

These are design decisions and platform realities, not bugs; each has a recommended posture. Read this page before you promise coverage to your auditors.

Platform

1. macOS only, today. Windows support is an active workstream but not yet shipped. iOS and Android are out of scope entirely — on mobile, the operating system closes all three of our enforcement points by design (no system-wide clipboard hook, no app-supervised local TLS proxy, no browser extensions with the required reach). Do not plan mobile coverage around this product.
2. Input Monitoring is a per-machine human step. No MDM or profile can grant it (see Enrolling devices). Until granted, that device runs reduced and reports itself degraded.

Trust and tamper model

3. Statuses are self-reported. Proxy health, CA trust, and agent version come from the agent itself. The console renders staleness honestly (Unknown, never a stale green), but a technically capable insider who can extract the agent’s credentials could replay false status. The posture: honest against casual tampering; devices that must resist a determined insider belong on MDM-enforced trust and tighter MDM controls.
4. Local (Tier L) trust is per-user and employee-revocable. On a Mac with multiple user accounts, trust approved by one user covers only that user’s session — a second account’s traffic is uninspected while the device row reflects the reporting user’s state. If you run shared or multi-user Macs, use the CA certificate profile.
5. “Was trusted, now isn’t” alerting is coming, not live. Today a device whose trust was quietly removed shows its current honest state — the console does not yet raise a distinct tampered alert for the transition. Until it does, treat unexpected regressions in the CA trust column as worth investigating.
6. The agent does not resist a malicious local admin. Self-protection against a user with root on their own machine is explicitly out of scope; MDM-level controls are the right layer for that.

Interception coverage

7. Certificate-pinning apps defeat the proxy — by design of TLS. The browser extension covers the pinned-browser case; a pinned native app’s traffic is not inspectable (its uploads may still be caught at copy/paste time by the clipboard sensor).
8. The proxy blocks; it does not redact in flight. Redaction happens at paste/submit. An upload either goes through or it doesn’t.
9. Interception scope = the platform catalog. Domains outside the catalog are not intercepted at the network layer — a containment feature, since the device certificate is constrained to the catalog (see How it works). Device-wide copy-time rules still apply everywhere. Missing a destination you care about? Contact support to propose a catalog addition.
10. Safari extension enablement needs DDM. A supervised Mac on macOS 15+ and an MDM that supports declarative device management. A profile-only MDM applies nothing — silently. Chromium force-install requires the extension to be configured for your deployment.

Operations

11. No MDM API integration, deliberately. The console generates profiles; a human pushes them through your MDM. We never hold credentials that could push trust anchors or software to your whole fleet — that credential would be a bigger risk than the toil it saves. The cost: per-device CA profiles mean N downloads for N devices when rolling out MDM-enforced trust. Plan for it, or keep most of the fleet on Local trust.
12. Every device has its own CA; profiles are not interchangeable. A reinstalled device mints a new CA, silently orphaning any previously pushed trust profile — after a reinstall, re-download and re-push that device’s CA certificate profile.
13. Offline devices read as Unknown. A Mac that hasn’t checked in shows Unknown across its status columns until it returns. Unknown means “not currently accounted for” — actionable information, not noise.
14. Enrollment credentials are single-use and shown once. A lost install link or token is revoked and reissued, never recovered.