AI coach disclosure & safety

You're talking to an AI, not a person

The Pulsus coach is an AI assistant. Every message you receive from it is generated by a large-language model operated by us through a third-party inference provider under a zero- retention contract. We label the coach as AI in-product, in our onboarding, and right here for the record.

If at any point you would prefer to talk to a human, stop using the coach and contact our human support at support@pulsus.co, or — if you are in crisis — use the emergency numbers listed in our medical disclaimer.

What the AI coach remembers

• Session memory (hot): your current chat thread. Cleared when you close the thread or after 7 days of inactivity.
• Rolling digest (warm): a 90-day rolling summary of your previous chats + key HRV/stress readings. This is what lets the coach say “on Tuesday you mentioned…”. Overwritten continuously; older entries fall off as new ones land.
• Long-term profile (cold): an aggregated coherence profile summarising tendencies across months (no verbatim chat content).

You can clear all three tiers at any time from Settings → AI Memory. A clear is immediate and irreversible.

Automated decisions (GDPR Article 22)

Two pieces of output from Pulsus qualify as profiling under GDPR Art. 22:

• Your stress score (0–100) is computed algorithmically from RMSSD / SDNN / LF-HF ratio against a rolling personal baseline. It is not used to make legal or similarly significant decisions — only to tune the coach's suggestions and ritual recommendations.
• Your persona archetype (stressed achiever, optimizer, calm seeker, burnt out) is inferred from your onboarding answers. It influences the default coach voice and the framing of suggestions, nothing more.

You have the right to human review of either classification. Email privacy@pulsus.co with “Art. 22 review” in the subject; a human on our team will respond within 14 days.

Crisis routing

The AI coach applies a two-stage crisis filter to every inbound message: a fast keyword pass + a foundational-model classifier. When the filter flags a message as indicating self-harm, active crisis, or acute medical distress, the coach pauses its normal response and:

1. Returns a short, non-clinical message acknowledging what you shared.
2. Surfaces country-specific emergency resources based on the locale we have on file (988 US / 111 UK / 112 EU / 182 TR / 13 11 14 AU / Yorisoi JP).
3. Offers the option to connect with a human (support email), and a reminder that the app is not a crisis service.

This is a safety heuristic, not a clinical triage. If you need help, please call the numbers above — they are staffed by real people.

Training data + your messages

Your chat messages are used only to produce the coach's reply and to build the 90-day rolling digest described above. We do not use them to train any public LLM, and our third-party inference contract explicitly excludes training use. We do not sell chat content. We do not share it with advertisers.

AI-composite testimonials

Some testimonials on our marketing surfaces use AI-generated composite portraits rather than photos of real users. When this is the case we label the card with an “AI composite” tag per the US Federal Trade Commission's 2026 endorsement guidance. The underlying narrative is still based on actual pilot data.

Scope + limitations

• The coach is not a therapist, counsellor, psychologist or medical professional.
• It can make mistakes. Please verify any suggestion that would affect your sleep, medication, or medical routine with a real clinician.
• It does not continuously monitor you. It only knows what you tell it and the metrics we have on file.

Questions? Email privacy@pulsus.co.