Trust Charter

Effective 2026-11-07. This is the plain-language summary for Krewva's iOS, Android, macOS, and Windows applications. The formal Privacy Policy controls if this Charter differs.

Krewva earns trust by showing what it sees, what stays on the device, what may reach the provider for the feature you choose, and what authority remains with you.

Mobile Reply provider notice updated 2026-11-07. Reply requires its separate 2026-10-22 acknowledgement. The general Terms and Privacy acceptance remains version 2026-09-04.

What Krewva will do

Krewva will keep permissions and authority separate.

Operating-system screen access, feature enablement, AI-processing consent, Task acceptance, Step approval, and outbound action authority are different decisions. A Task candidate is not an accepted Task. A Task is not a Step approval. A suggestion is not a sent message.

Krewva will use the smallest screen view the feature needs.

On macOS and Windows, capture is user-triggered and limited to the active top-level window. It is not continuous monitoring and never falls back to a full-desktop capture. Denied, revoked, protected, or changed targets fail closed. You are responsible for lawful access and any notice or consent required for other people visible in a window.

Krewva will explain three different desktop screen flows.

Draft a Reply may send one fresh active-window screenshot through Krewva to a third-party LLM provider after feature enablement and AI-processing consent, then show reviewable suggestions; the capture itself never sends. Do a Task local OCR tries semantic text first, uses bounded local OCR only when needed, and keeps raw pixels on the device; selected derived text may support a Task. Do a Task screenshot discovery is a separate enabled and AI-consented upload to a third-party LLM provider for Task candidates only. Task acceptance and Step approval remain separate.

You choose what to insert on iOS and Android.

Reply on iOS and Android uses a third-party LLM provider. Before a Reply request is sent, Krewva asks for separate Reply permission. The request may include the current conversation text or screenshot, your reply instructions, and selected personalization context, such as relevant profile, contact and writing information. You review the drafts and choose what to insert. Krewva never sends a message for you. Basic typing does not itself start a Reply request.

Other AI features keep their existing permissions.

Voice Input, Dictate, Intent, Refine and independent personalization learning use third-party LLM providers under their existing permissions. Desktop AI processing is unchanged. Agreeing to the general AI notice alone does not grant Reply permission. Turning off AI drafts revokes the general AI and Reply permissions; Voice Input has its separate permission.

Krewva will keep raw screenshots out of ordinary history.

Screenshots captured for AI features are request-scoped and excluded from Krewva account history, ordinary logs, Sentry, and support bundles. Krewva may keep permitted derivatives such as drafts, review evidence, selected Task evidence, plans, outcomes, audit facts, or a one-way fingerprint under their separate rules.

Feedback includes only what you choose to send.

Feedback text and selected screenshots go to the Krewva support team in a private Slack channel, with your account ID and app version. They follow the support workspace's access and retention settings. Krewva account deletion does not automatically delete Slack feedback.

Krewva will describe its providers' data rules.

For the third-party LLM features described above, API data is not used to train or improve the providers' models unless Krewva explicitly opts in. Default abuse-monitoring logs may be kept for up to 30 days. Krewva does not promise zero provider retention. Cloudflare handles website hosting, waitlist processing, KV storage, and rate limiting. Final dictation text may pass through Krewva and a third-party LLM provider once for faithful cleanup, without server-side transcript history.

Reply has provider-specific data rules.

Reply is subject to the provider terms and retention rules described in the sub-processors list. Krewva does not promise zero provider retention or US-only processing.

Mobile saved context can leave the device through backup.

On iPhone and Android, saved history and personalization are stored on your device and can be backed up automatically to your Krewva account when signed in and connected. Android writing learning starts on for new local profiles. Original text typed with Krewva or dictated may become saved Writing Samples; unchanged AI replies and Intent output are excluded from original-writing samples. Saved samples can leave the device through backup even when basic typing does not start a Reply request. Selected saved context may also go to an AI provider after the relevant consent.

Mobile analytics and crash reporting have separate purposes.

Core app and keyboard-quality analytics are enabled after legal onboarding when collection is available and the account is eligible. Fixed events use an account-linked identifier and go through Krewva to PostHog's US endpoint. This path excludes conversation text, screenshots, audio and keystrokes. Configured Sentry reporting sends bounded error and crash diagnostics separately. The Privacy Policy explains controls, retention and deletion requests.

Krewva will keep retention claims specific.

Desktop cached message bodies, cards and voice samples are purged after 365 days of account inactivity. Mobile saved history and personalization remain until you remove them. Contact learning has separate pause and resume controls; forgetting an Android contact item does not itself pause future learning. Backup changes finish when connected. In-app account deletion on iPhone and Android starts immediately. Consent-gated redacted improvement records and audit records have separate lifecycles; they are not raw message history.

What Krewva will not do

Krewva will not monitor the whole desktop.

It will not continuously watch the screen, silently fall back to a full-desktop screenshot, or continue with a denied, revoked, protected, or changed target.

Krewva will not turn a capture into authority.

A captured window can produce reviewable suggestions or Task candidates. Capture does not send, Task candidates do not execute, Task acceptance does not approve a Step, and no outbound action happens without the authority that action requires.

Krewva will not describe a provider promise as absolute.

A provider's default API no-training position does not mean abuse monitoring is absent. Krewva will not promise zero retention or hide that limitation.

Krewva will not call permitted derivatives raw history.

A draft, selected Task evidence, audit fact, or redacted improvement record may have its own lifecycle. That does not make it a stored raw screenshot or raw message archive.

If Krewva fails you

A permission or consent check fails.

Krewva keeps the relevant gate visible and fails the operation closed. It does not silently substitute a broader screen, a new permission, or an unapproved provider.

A screen target changes or is protected.

Krewva stops the capture or processing path. The user can retry only after the feature can safely bind to a new permitted active window.

You want to leave.

In the iPhone or Android app, use Settings → Account deletion. The accepted request signs you out and starts the automatic purge immediately; a completion notice follows. Older desktop clients may still use a cancelable 30-day scheduled deletion. Third-party revocation is best effort, and legally required or separately governed records may remain for their stated lifecycle. You can also request deletion without opening the app by emailing privacy@krewva.com.

You have a privacy question.

Contact privacy@krewva.com. The formal Privacy Policy controls the legal meaning, and the public sub-processors list identifies the current operational providers.