Trust
AI & Safety
OwnCadence uses AI to support a personal productivity or routine goal. This page explains what the system does, what it must not be used for, and the controls that keep decisions with you.
Know this before you use the mentor
- You are interacting with an AI system, not a human coach or professional adviser.
- AI can propose and explain; it cannot directly change your canonical plan without product validation and your decision.
- The service is limited to productivity and routines—not medicine, therapy, nutrition treatment, crisis care, or emergencies.
- If there is immediate danger, call 112 in the EU or your local emergency number; do not wait for OwnCadence.
1. You are interacting with AI
The product identifies the mentor as AI at or before the first interaction and keeps that identity visible in the experience. There is no human silently composing the replies, and support staff do not watch mentor interactions in real time.
AI can sound confident even when it is wrong. A fluent explanation is not proof that a statement is factual, safe, suitable, or based on complete context. Use your judgment and verify anything important.
2. What the AI may help with
- turn a goal and constraints into a small, bounded weekly routine;
- explain the reason, timing, fallback, or expected effort for an action;
- summarize check-ins and observed completion patterns;
- propose a smaller, clearer, or differently timed next-week plan;
- suggest a memory candidate that you can inspect before it is used; and
- help you reflect on what was workable without pretending to know your inner state.
The AI receives a task-specific, minimized context assembled by the OwnCadence server. It does not independently browse your device, read email or calendars, contact people, make purchases, access health records, or execute work on your behalf. Those capabilities are not part of the current service.
3. How you stay in control
| Stage | System role | Your control |
|---|---|---|
| Goal and constraints | OwnCadence structures what you enter and checks required fields. | You define and confirm the goal, limits, schedule, and acceptable fallback. |
| Weekly program | AI may draft; deterministic rules enforce bounded duration, valid actions, and safety constraints. | You can edit before confirmation and can later revisit the goal or pause it. |
| Adaptation | The system links a proposal to observed evidence and shows the exact before/after difference. | You can accept, edit the next week, reject the proposal, or keep the current plan. |
| Memory | The system distinguishes explicit facts, observations, and proposed inferences. | You can inspect, correct, dispute, stop using, or delete memory; optional sensitive context has a separate control. |
AI output never receives direct authority to write canonical account state or call an external provider. Product-owned validation and authorization checks remain between a generated response and any durable change. If AI is unavailable or rejected by safety/schema checks, the product can use a deterministic fallback or ask you to try again.
4. Health, professional, and high-impact boundaries
OwnCadence is designed for ordinary productivity and routine support. References to sleep, food, exercise, mood, or family are accepted only as limited context for feasibility. They do not turn the service into healthcare or professional advice.
- No diagnosis, treatment, prevention, symptom assessment, medication advice, or clinical monitoring.
- No psychotherapy, counselling, suicide-risk assessment, crisis intervention, or substitute for a qualified mental-health professional.
- No diet plan, calorie target, eating-disorder guidance, supplement instruction, or therapeutic nutrition advice.
- No legal, financial, employment, insurance, education-admission, housing, or essential-service decision support about another person.
- No autonomous decision with legal or similarly significant effect about you.
- No biometric categorization, facial analysis, or emotion-recognition system.
If a request crosses these boundaries, the mentor should decline that part, explain the limit, and redirect to an ordinary productivity-safe alternative or an appropriate qualified resource. A refusal is a safeguard, not a professional evaluation of your situation.
5. Urgent safety and crisis situations
- If you or another person may be in immediate danger in the European Union, call 112 now.
- Outside the EU, call the local emergency number or go to the nearest emergency department.
- If it is safe, contact a trusted person who can stay with you and a qualified local crisis or healthcare service.
- Do not wait for an OwnCadence response or email reply before seeking urgent help.
OwnCadence uses a product-owned safety boundary that can interrupt routine coaching when text appears to indicate acute self-harm, violence, medical, or other immediate risk and show a reviewed resource. Automated classification has false positives and false negatives. It does not assess, diagnose, or replace local professionals.
- 112 — the EU emergency number
Official European Commission information.
6. Memory and optional sensitive context
OwnCadence memory is a visible product feature, not hidden psychological profiling. It separates: (a) facts you explicitly state and save; (b) observations from actions or check-ins; and (c) inferences proposed by the system. Proposed inferences do not become unquestionable truth and should remain traceable to evidence.
Optional context about sleep, food, exercise, mood, or family circumstances is off by default and requires a separate explicit choice before it is remembered for personalization. You can use the core service without it. You can withdraw the choice, correct the context, stop it from being used, or delete it. Do not submit detailed health records or another person's sensitive information.
A model can infer more than you intended from ordinary language. We limit the context sent to the requested feature, reject unregistered or unsafe output categories, and keep general analytics and operational logs free of raw mentor content by design. See the Privacy notice for lawful bases, recipients, retention, and rights.
7. AI provider and data handling
The planned production path uses OpenAI's Responses API through an OwnCadence server adapter when AI is enabled. We send the minimum context needed for the specific task and structured output requirements. OpenAI does not receive account credentials or direct database access.
OpenAI states that business API inputs and outputs are not used to train its general models by default. Standard API controls may retain content for up to 30 days for abuse monitoring and may retain some Responses API application state for up to 30 days. OwnCadence does not opt in to provider model training with account content. Any materially different provider, training choice, regional setup, or retention position requires a new review and updated notice before activation.
- OwnCadence Privacy notice
Full data-use and transfer information.
- OpenAI API data controls
Current provider documentation.
8. Known limitations and safer use
- Missing context: the model sees only a selected slice of your information and may not know what changed offline.
- Hallucination: it can invent a fact, explanation, or source, or combine details incorrectly.
- Bias and mismatch: suggestions may reflect patterns that do not fit your culture, disability, resources, work, or family context.
- Recency: model knowledge and provider behavior can change, and a generated statement may be outdated.
- Automation error: validation and safety checks reduce risk but cannot eliminate every incorrect or unsafe output.
- Availability: provider, network, or product failures may delay or prevent an AI response.
For safer use, keep goals ordinary and bounded, provide only necessary context, read the reason and exact change, reject anything that does not fit, avoid acting on surprising factual claims without checking them, and seek a qualified person when the issue belongs to a professional domain.
9. Testing, monitoring, and changes
Before a real model or model version is enabled, OwnCadence requires schema validation, safety tests, regression evaluation, and a controlled activation path. Production controls can disable the AI path, reject malformed or policy-bypassing output, and fall back without granting the provider authority over canonical state.
We monitor content-free reliability and safety outcomes rather than storing raw mentor conversations in general logs by default. A support or incident report you send may be reviewed by authorized people. We may update models, prompts, validators, or provider configuration; a material change to capability, data use, risk, or user control requires review and an updated disclosure.
The service is for adults aged 18 and over. We do not design or evaluate this mentor for children, and an adult should not create an account for a child.
10. Report a safety concern
If the mentor produced an unsafe, discriminatory, boundary-crossing, or misleading response, stop following it and email hello@owncadence.com. Include the approximate time, feature, language, what concerned you, and a screenshot only if you are comfortable sharing it. Remove unrelated personal or third-party information.
Email is for product safety and support, not emergencies, and is not monitored continuously. We use reports to investigate the feature, contain risk, and improve tests or controls. We will handle report data as described in the Privacy notice.