How we use AI

Connections between saved check-ins

If you enable “Connect my saved check-ins” in your Personal Map, OpenAI can use summaries from up to six recent saved moments and follow-ups to suggest a shared pattern, helpful condition or possible underlying need after at least two distinct experiences. This feature does not read raw past-visit history. Saved AI summaries can be mistaken; they are not treated as independent confirmation of a pattern or a hidden truth.

Connections update when you return to your dashboard or Personal Map after saving new experiences. Editing or deleting a source clears the previous connections. You can inspect their sources, correct the saved moments, or turn off and delete connections in your Personal Map. They are included in your account export and removed with your account. Requests already sent to OpenAI cannot be recalled. If no clear connection is supported, or AI is unavailable, your individual saved observations remain available.

Automatic Personal Map updates

Saving a moment carries its displayed conditions, patterns and desired direction into your map. “Update my Personal Map” sends your follow-up, the saved moment and up to five earlier follow-ups from that same map to OpenAI when available, then saves the generated notes automatically. You can correct or delete them. They are tentative interpretations, not diagnoses or confirmed facts. Raw follow-up writing is not stored. If AI is unavailable, a labeled basic outcome note is saved without inventing conditions or desires.

Guided reflections

The Guided Moment can use a bounded AI task to reflect only what a person chose to write. Before anything is sent to OpenAI, the person must say the moment feels safe enough to explore here and explicitly agree to the reflection and proposed plan. A reviewed non-AI option is always available. If live AI is unavailable, disabled, or rejects the request, the same flow continues with a human-reviewed deterministic reflection. AI does not determine safety, infer a biological nervous-system state, diagnose a condition, or invent a support intervention.

Personal plans and guided questions

A plan-enabled task can propose a goal and personalize the titles, explanations and optional questions for a defined set of practical activities. It uses the current check-in and, when account history is enabled for this request, bounded relevant past writing. It does not search unrelated conversations, speak as your unconscious or claim to uncover hidden motives. You can correct the result or begin from your own words.

If you choose Update reflection and support plan, OpenAI may process your original check-in together with the added context you enter. Typing a correction does not send it to AI. If an update is unavailable, your words stay available to edit or retry.

Starting a plan explicitly keeps the displayed reflection, goal and module framing in your account. Your chosen practical steps are kept as you continue. If personalized AI is unavailable, the app identifies its general starting point instead.

Relevant past context

Where history is available and you enable its use, a check-in request can include up to three relevant dated reports from your retained check-ins. We use your writing, with newer corrections taking priority. We do not treat earlier AI suggestions as evidence, intentions as completed actions, or earlier safety choices as proof of present safety.

You can inspect which past writing was used, correct or exclude it, or begin without past context. Saving this visit and using past writing are separate choices. If a source changes during a request, the response falls back to current support. Already-sent provider requests cannot be recalled. No provider-managed conversation memory is created.

Guided module conversations

Choosing Continue with AI sends the saved reflection and goal for this plan, its module guidance, and the replies in your current module to OpenAI. The conversation can offer a tentative perspective and one relevant question at a time. You can disagree, correct it, or continue with prepared guidance without writing. It cannot know hidden causes or replace professional support.

Conversation replies remain in the open module and are not saved to your account or the AI usage ledger. Closing or refreshing the module clears the conversation. When you continue from the practical module’s closing page, the displayed next step is kept in your plan. You can adjust it first. If AI is unavailable, the app labels its prepared guidance and does not claim to have interpreted your reply.

Bounded roles

An approved AI task may reflect user-provided words, offer one grounded perspective to consider, ask one non-leading question, notice conditions the person described, or summarize a reported outcome. It may later help rephrase approved content, draft a smallest next action from structured choices, draft a support request, or summarize user-approved patterns. Every approved task must have source content, validation, a deterministic fallback, and a way for the user to reject or edit the result.

Prohibited roles

• Diagnosing or naming a disorder.
• Determining whether a person is safe or in crisis.
• Inferring trauma, dissociation, or a precise autonomic state.
• Claiming privileged access to a person's motives or inner truth.
• Inventing an intervention or contradicting an exclusion.
• Telling a person to change medication or professional care.
• Automatically contacting another person.

Data and accountability

Every AI feature must use the minimum necessary information, remove direct identifiers where feasible, validate structured output, record its source and version, and offer an approved deterministic fallback. Raw prompt and response text are not stored in the AI usage ledger. The provider request sets the OpenAI API's store option to false. User data will not be used for model training without separate, explicit, revocable consent and approved contracts.