Your support. Your choices.
Settings & Controls
Choose Call Shield™ warnings, family help, and what your family can decide. Your parent always decides whether to keep talking, hang up, or call family.
Product review · Proposed v0.1 · October 29, 2026
Explore the controls and read the complete rules here. This preview does not place calls, send alerts, or save personal settings.
Sample profile · Every service starts Off
Preview only · No personal data or permissions are saved.
Service settings
Call Shield™
Plain-language warnings during suspicious calls.
The complete product rules
One place to review it all.
The control-center direction is reflected below. Defaults, timers, permissions, and exact wording remain proposed for your review.
Navigation and screen structure
Use a standard gear icon beside “Settings” in the app header and family dashboard. The destination title is “Settings & Controls”; never rely on the icon alone. Guardian View™ opens the same controls through “Manage settings” for an explicitly selected person.
Keep “Settings for [name]” visible on every screen. Mom, Dad, and other supported adults have separate profiles, consent, contacts, and permissions. Switching profiles must not copy settings.
| Section | Contents |
|---|---|
| Services | Call Shield™ and Call Family™ rows and their detail screens. |
| Who helps & backups | AI support, family help, and approved backup contacts. The parent always controls the call. |
| Family permissions | People, service scopes, allowed decisions, approval rules, and access expiry. |
| Privacy & sharing | Separate AI and family permissions; conversation privacy and revocation. |
| Display & notifications | Caption readability, text size, and notification preferences. |
| Activity & changes | Help requests, delivery outcomes, approvals, and who changed each setting. |
Each service row shows its icon, name, short description, On/Off switch, operational status, and “Who helps: [choice].” Tapping the row opens details; tapping its switch changes only enablement. Detail order: service switch → Who helps → contacts and backups → service controls → permission summary.
Starting state
For a new profile, Call Shield™ and Call Family™ are Off. “Set up” gathers the required permissions, contacts, and configuration; the user reviews the proposed route and taps “Turn on [service].” Nothing is enabled merely by accepting an invitation. Existing valid choices are preserved during migration; new grants and new sharing stay Off.
Captions and Call for Help are persistent interface controls, not service switches. Their exceptions are defined under “Defaults and safety exceptions”. There is no “Turn everything off” control in this version.
Service inventory and controls
All service switches start Off. Defaults below apply inside setup and become effective only after confirmation. AI = “AI helps me”; Family = “Family helps me”; Both = “AI + family.” A suggested route never silently selects recipients or grants access.
| Service and stable key | Controls and proposed defaults | Route and boundary |
|---|---|---|
| Call Shield™ call_shield | Scam warnings: On with the service. Spoken warnings: Off. “Automatically ask for help”: Off; concerns still appear on screen. | Suggest Both. All routes supported. Routing chooses who responds to a concern; consented scam detection remains part of the service. |
| Call Family™ call_family | Choose and order contacts. Each call or live-call invitation is initiated visibly by the user or an approved help request. | Suggest Family; Family or Both. AI may assist while connecting. AI alone is unavailable: “This action needs a family contact.” No automatic or silent join. |
Live-call joining requires Call Family™ and the parent’s visible per-call approval. Setup asks before enabling a dependency. Turning a service Off does not end an active call. Companion™ never takes over the call; your parent always decides whether to keep talking, hang up, or call family.
Routing and escalation
| Who helps | Exact helper text and behavior |
|---|---|
| AI helps me | AI points out warning signs and helps the parent check a request. AI does not speak to the caller or take over the conversation. The parent controls the call. |
| Family helps me | Ask the chosen family member to join with the parent’s per-call approval. If nobody answers, the parent decides whether to keep talking, hang up, or try another contact. |
| AI + family | AI points out warning signs while the parent invites family. Family joins only with approval. An unanswered invitation never transfers control of the call to AI. |
Use one three-option radio group, not three independent switches. Unavailable modes remain visible with their reason. Family/Both requires at least one accepted, permitted contact. AI/Both requires current AI consent and supported capability. Choosing family help does not disable separately consented scam warnings.
When an event starts
A manual request starts when the parent asks for help. A Call Shield™ concern starts an automatic help request only when “Automatically ask for help” is On. A help request does not admit anyone to a live call without the parent’s per-call approval.
| Event class | Contact sequence and no-answer behavior |
|---|---|
| Requested help during a suspicious call | Offer contact 1 at t=0; contact 2 at +30 sec; contact 3 at +60 sec. Exhaust at +90 sec, or 30 sec after the last available contact. If nobody answers, the parent decides whether to keep talking, hang up, or try again. These are proposed notification timers. |
Use up to three ordered contacts. Skip declined, revoked, unavailable, or failed-delivery contacts immediately. Delivery without “I can help” is unanswered. Offer each contact once per request; a manual retry creates a linked attempt. AI never takes over an unanswered call.
The first valid “I can help” temporarily claims the help request; expired or later offers cannot claim it. Acceptance pauses contact offers but does not mean “Resolved” or “Joined.” If joining fails, the responder declines, or no connection occurs within the proposed 60-second interval, offer the next approved contact. After exhaustion show “No one has answered,” “Try again,” and “Call for Help”; leave the request unresolved. The parent controls the call throughout.
AI failure. Show “AI is unavailable” and keep manual family help available. No mode change without consent. Neither an AI failure nor an unanswered family request transfers control of the call.
Off stops new triggers; it does not hang up a call or cancel an active help request. Show “Service off · Help request still active” and “Review request.” Explicit cancellation stops unsent follow-ups. Consent/access revocation blocks affected actions immediately. Never infer safety from silence, timeout, delivery, or reassurance.
Family decision permissions
The supported adult is the profile owner. An adult child, spouse, sibling, caregiver, or payer receives no decision rights by relationship or payment. An accepted invitation creates a contact, not an administrator. Version 0.1 delegates decisions to adults only. Any legally authorized representative requires a separate verified authority workflow; a family toggle or unanswered alert never establishes incapacity.
For each named person, show “What can [name] do?” Permissions are scoped to one profile, service, action, situation, and optional expiry. All grants start Off. Each enabled decision grant uses “Ask me first” by default; the owner may select “May decide” only for the bounded actions below.
| Permission label | Allowed scope and limits |
|---|---|
| Receive alerts | Select call concerns and help requests. Basic alerts contain the request category and time. Alerts do not grant access to a conversation. |
| See updates in Guardian View | Choose call concerns and family help status. Reading an update is separate from acting on it. |
| Help with a request | Accept or decline an offered request, call back, and record “Followed up.” Only the owner or a scoped responder may close it with an explicit outcome; no “Safe” badge is inferred. |
| Join a call when invited | Join only the named, active request after the owner’s visible per-call approval. No background listening or silent patch-in. Existing permission does not itself admit the person. |
| Change selected services or who helps | May decide can cover support routing within already-consented AI and contact choices. Disabling Call Shield™ always asks the owner first. These permissions never grant control over the call. |
| Choose a backup contact | May decide can reorder only contacts already approved for that service and data scope. Adding a person or expanding their information access requires the owner. |
Situations and approval
Choose “For routine tasks,” “When I ask for help,” and/or “When I do not respond.” No response means the owner, never an unanswered family alert. The proposed owner-prompt interval is 30 seconds during a suspicious call while help remains available. This condition never broadens a grant or allows AI to take over. Preview it: “Sarah may reorder Eleanor’s approved backup contacts during a help request.”
An “Ask me first” change stays “Waiting for [name]” and has no effect until approved. Routine proposals expire after 24 hours. Event proposals expire when the event ends or its next contact interval expires; they must not delay help or emergency access. Owner decisions take priority; stale or conflicting edits require review.
Only the owner or verified representative may grant/revoke access, broaden sharing, add recipients, change account ownership, or approve sensitive changes; require reauthentication for these changes. Family cannot enlarge its authority. Revocation blocks queued actions and access immediately. Notify the owner of applied family changes and log actor, scope, before/after, time, and approval. Protected Off changes show “Turn off [service]?” with the impact, “Turn off,” and “Keep on.”
Defaults and safety exceptions
| Control | Proposed rule |
|---|---|
| Captions | Show “Captions — Always on” as a status, with text-size/contrast controls. Do not add a normal Off switch. Captions operate only within an authorized audio session; privacy consent is not bypassed. Show “Waiting for audio” or “Captions unavailable” honestly. |
| Call for Help | Visible on every screen, including setup, Off, and degraded states. Open “Call my family” and “Call emergency services.” Use the validated local emergency number and device calling flow. No account login, AI reply, or family approval may delay access. |
| Urgent danger | Present emergency calling immediately; do not wait for the contact sequence. AI can support communication but cannot dispatch emergency services or promise a response. This version has no automatic emergency call or professional monitoring. |
| Call Shield boundaries | Warn and offer the user choices. Neither AI nor a family grant may automatically hang up, seize a call, block transactions, move money, or authorize a payment. The user controls the call. |
| Notifications | Quiet hours: Off until set. They suppress routine summaries, not user-requested help. OS notification/DND limits remain visible. |
| Connectivity and availability | Show what is unavailable and the last successful update. No offline remote-setting saves or “Family notified” without delivery evidence. Keep available device calling options. Unsupported telephony or captioning shows “Not available on this device.” |
| Conversation privacy | Companion™ never sells your conversations or uses them for advertising. AI processing and family sharing have separate consent. Raw call recording/transcript storage is not offered in v0.1. |
Designer copy and states
| State | Label and required behavior |
|---|---|
| Disabled / incomplete | “Off” / “Set up [service]” / “Needs setup.” Explain missing consent, recipient, or device permission beside “Finish setup.” |
| Saving / pending / failed | “Saving…”; “Waiting for [name]”; “Couldn’t save. Your previous settings are still active.” Never show On before the server confirms the change. |
| Enabled / degraded | “On” with “Who helps: …”; “Needs attention — [reason]” if a dependency fails. Keep intended On separate from operational availability. |
| Active help | “AI is helping”; “Contacting [name]”; “Delivered to [name]”; “[name] accepted”; “[name] joined”; “No one has answered.” Each requires its own evidence. |
| Access / closure | “View only”; “Ask [name] to change this”; “Permission expired”; “Cancelled by [name]”; “Closed by [name] — [outcome].” Never auto-label a person “Safe.” |
Use visible On/Off text, large type, at least 48 × 48 logical-pixel touch targets, keyboard and screen-reader names, and focus-visible states. Do not rely on color. Plain “AI” means the artificial-intelligence assistant; introduce that meaning in setup. Use “Mom or Dad” in family marketing, and the person’s name in the app.
Engineering contract
| Record | Minimum fields |
|---|---|
| ServiceSetting | profile_id, service_key, enabled, route {ai,family,both}, ordered_contact_ids, fallback_to_ai, fallback_to_family, trigger settings, consent_refs, expected_revision. |
| PermissionGrant | owner_profile_id, grantee_id, service_keys, action_keys, data_scopes, situations, approval_mode {ask_first,may_decide}, expires_at, revoked_at. No wildcard grants by default. |
| HelpEvent | event_id, profile_id, service_key, trigger, urgency, policy_version, route snapshot, opened_at, deadlines, per-contact delivery/acceptance/join status, current responder, outcome and closing actor. |
| Change and consent log | actor, profile, affected setting/grant, before/after, approval actor, timestamps, revision. Keep event/change metadata for 90 days as a proposed default; exclude raw audio, message bodies, other private content. |
Authorize every read, write, dispatch, and join on the server using current consent, grants, expiry, availability, and situation. Revocation overrides an event’s route snapshot. Use compare-and-swap revisions and atomic dependency updates; on conflict reload for review. Log changes while redacting sensitive field values.
Persist deadlines across restarts. Deduplicate dispatch by event + recipient + attempt and repeated detector signals by open event. Track Open, Contacting/AI helping, Accepted/Connected, and explicit Closed/Cancelled. “No answer” stays unresolved. Send only currently permitted data.
Acceptance criteria
| Test | Required result |
|---|---|
| New or migrated profile | Call Shield™ and Call Family™ rows only; no new consent or family grant inferred. |
| Both and no answer | AI points out warning signs and approved family contact offers advance once. After exhaustion the help request stays unresolved. The parent decides whether to keep talking, hang up, or try another contact; AI never takes over. |
| Family fallback | An unanswered family request offers another trusted contact or a retry. It never gives AI control of the conversation. |
| Permission boundaries | A child’s grant for Dad cannot affect Mom. A permission to receive a call concern cannot broaden sharing, disable protected settings, or transfer control of the call. |
| Revocation and races | Revoke before queued dispatch, accept, or join: the action fails closed. Simultaneous setting edits cannot overwrite a newer owner choice. |
| Dependencies and failure | Disabling Call Family™ shows the affected routes. Declining the change preserves the prior configuration. Offline or rejected saves never display a false On. |
| Persistent controls | Captions have the specified status/activation behavior; Call for Help remains reachable with all services Off and during outages. Emergency access never waits for AI. |
| Truthful outcomes | Delivery ≠ accepted ≠ joined ≠ resolved. “No response” never means a safe call or person. Revoked recipients lose access immediately. |
Remaining validation
Before production, validate devices/live-call joining, caption activation for any certified lane, detector-trigger quality, and proposed timing. These are release dependencies, not current-support claims.
Family settings lock — additional rule
A settings lock starts Off. The profile owner must explicitly authorize a named adult, selected services, and an expiry before the family member may lock those settings. Permission to receive alerts or help with a request does not include permission to lock.
A lock prevents accidental edits; it does not grant new decision rights or authority over the supported adult. The owner can reauthenticate to unlock or revoke it. A verified representative needs the separate authority workflow. Call for Help, ending a call, and emergency access remain available.
Show “Locked by [name],” “Ask [name] to change this,” and “Unlock settings.” Real locks require server authorization, rate-limited credential checks, audit history, and owner recovery. The 1234 code in the interactive family-lock example is a public demo code only.