Message Content Policy

Vireo uses automated tone checks to help with co-parent communication. This page explains the checks in direct-message composers, email-style compose and reply screens, and daily check-in notes and replies.

A tone block in these screens leaves the draft available to edit and explains the result. A blocked draft is not sent by that send attempt.

These are writing aids in the app. They do not guarantee that every harmful message will be detected or prevented, and they do not apply to every kind of content, such as photos or uploaded documents.

Local checks look for words, phrases and text patterns, including insults, threats, guilt or pressure, and confrontational wording. In message and email screens they can run while you type and when you try to send. Check-in notes and replies are checked when you try to send. A matching phrase may trigger a result even when its surrounding context is harmless.

Direct-message and email drafts may also be sent to our AI processing provider before you send them. This includes drafts the local check did not flag, not just borderline messages. Some screens request analysis after a pause in typing; others request it when you tap Send. Drafts already classified locally as hostile or manipulative do not need that additional AI check. Check-in notes and replies currently use local checks. Our Privacy Policy identifies the AI provider and explains how message text is used.

Sent direct messages may also be analyzed afterward to store tone labels alongside the delivered message. That later analysis does not recall or unsend the message.

If AI analysis is unavailable, the local check remains available. Results can include a tone label, flagged wording and a suggested rewrite. Any court-risk label is an automated estimate, not legal advice or a prediction of a court's decision.

The check is not a judgment about who is right in a disagreement. It can miss harmful wording or flag harmless wording. Section 6 explains how to report a mistake.

The composer stops a send attempt when its current check classifies the draft as hostile or manipulative. Local rules include threats of harm, name-calling, some profanity, blanket accusations, custody or legal threats, and phrases that apply guilt or emotional pressure. AI analysis can also identify hostile or manipulative wording.

Some ordinary profanity is classified as hostile, even when it is not directed at your co-parent. Some phrases about custody or legal action can also trigger a block in a neutral discussion. The result depends on the rules and analysis, so these examples are not a complete list or a guarantee of how a particular draft will be classified.

The blocked draft stays in the compose box for you to edit. Opening this policy does not send it. If the classification seems wrong, follow the guidance in Section 6.

A warning or passive-aggressive classification is less restrictive than a hostile or manipulative classification. Direct messages and email screens show tone feedback, and these lower-level results can still be sent. The mobile email send check asks whether to edit or send anyway; web email shows feedback beside the draft without a separate confirmation.

The optional cool-down applies to flagged direct messages. It pauses sending for a few minutes. Editing the draft causes it to be checked again; a draft that remains flagged may start another pause. Emergency direct messages skip this pause but remain subject to the composer's hostile and manipulative checks.

Check-in notes and replies currently stop hostile or manipulative drafts. Lower-level results do not require a warning confirmation or cool-down in those screens.

Delivered communications and their available tone labels are separate from private draft-event logs. Direct-message flows and blocked check-ins attempt to record certain tone events, such as a blocked send or sending a warned direct message. Logging can fail, and email compose and reply do not currently record these draft events. This is not a complete record of every warning or attempt.

A saved draft-event log can include a timestamp, action, flagged wording and up to 4,000 characters of the draft. Its draft text is restricted to the author through the app's account permissions; it is not shared with the co-parent as a delivered message or included in the ordinary communication export.

If an authorized professional share includes messages, it may show aggregate counts of blocked attempts and messages sent after a warning for each parent. That shared summary does not include private draft-event text. Delivered messages, email and their exportable records have their own sharing rules described in the Privacy Policy.

Tone labels describe individual pieces of text. They are not a legal finding or a personal score.

If a normal message is blocked by mistake, review the flagged wording and suggested rewrite. You can edit the draft and try again without accepting the suggestion. Changing the wording does not guarantee a different result.

To report a mistake, email support@vireo.family with the screen you used, the explanation shown and a short example. Remove names, children's details and other private information that is not needed to explain the problem. A report does not automatically send or unblock the draft.

An unflagged result does not mean a message is safe, appropriate or legally advisable. The check is not human review, emergency assistance or a way to establish facts about a dispute.

The Privacy Policy explains how communication text, stored records and service providers are used. AI tone analysis can process unsent drafts as described above. Opening this policy from a draft does not include the draft or your account details in the policy link.

This page and its dated changelog describe changes to the checks and corrections to this explanation. An older installed app may show different controls until it is updated.

September 19, 2026 - Corrected the explanation of local phrase rules, AI processing of unsent drafts, differences between message, email and check-in screens, and author-only draft logs with shared aggregate counts. These are disclosure corrections; the moderation rules and permissions did not change.

August 22, 2026 - First public version of this policy.