AI Transparency
CharmGym uses AI as a practice partner—not as a person, therapist, matchmaker, or authority. This page explains what the preview does and where its limits are.
Last updated: September 2, 2026Every simulated conversation is visibly labeled as AI.
Feedback supports your own words instead of supplying a perfect line.
The coach refuses pressure, manipulation, and workarounds for consent.
1. What the preview AI does
The practice demo sends your current message, selected scenario, language, and “practice” mode to a server-side orchestrator. It returns a simulated character reply plus structured feedback across curiosity, reciprocity, clarity, authenticity, and boundaries.
The server-side orchestrator uses OpenAI’s API for the generated reply and keeps a deterministic local provider as a safe fallback. The external request uses a pinned production model, structured output, a short response limit, and disabled response storage. CharmGym never exposes the provider key to the browser.
2. What it does not do
- It does not know or contact the person you are practicing for.
- It does not predict attraction, compatibility, or real-life outcomes.
- It does not diagnose anxiety, attachment style, trauma, or any health condition.
- It does not replace professional support or emergency help.
- It does not remember your message across preview sessions or build a hidden personal profile.
3. Safety safeguards
Requests are validated before processing. A local rules layer and OpenAI moderation check input for crisis, coercion, sexual content involving minors, violence, and harassment. Generated output is moderated again before it reaches the browser. The coach responds with a boundary and safer next step rather than helping with harmful intent. Safety events contain category-level metadata, not the triggering conversation text.
These safeguards reduce risk but cannot catch every problem. Never treat an AI answer as permission to ignore another person’s words, discomfort, or refusal.
4. Quality and human responsibility
AI can be wrong. CharmGym validates response structure and tests common safety cases, but a fluent answer is not proof that the advice fits your situation. Use your judgment, choose low-pressure actions, and seek a qualified professional where appropriate.
5. Data and control
Preview practice text is processed by our Supabase Edge Function and OpenAI to produce and moderate the immediate response. CharmGym does not save it as a transcript, and API response storage is disabled. OpenAI does not use API inputs or outputs to train its models by default; it may retain abuse-monitoring data for up to 30 days unless a legal exception applies. See the Privacy Policy for details.
6. Report a problem
If an answer feels unsafe, manipulative, discriminatory, or simply wrong, stop following it and email help@charmgym.co. Do not include more personal conversation detail than needed. A dedicated in-product report control is planned before the mobile launch.
Ready to see the clearly labeled experience? Open the practice demo.