Yatri AI is designed as a bounded reflection tool, not a replacement relationship.

Section 01

What went wrong with Replika and Character.ai

Open-ended companion framing creates predictable boundary and safety questions that a user should be able to evaluate.

• Companion-style language can position software as a relationship unless the product repeatedly states the opposite.
• Long character role-play can make the tool boundary less visible, especially in age-sensitive contexts.
• Public product pages should be checked for explicit high-risk handling, memory controls, age gates, and emergency disclaimers rather than assuming those controls exist.
• Public product descriptions do not give a user enough evidence to assume clinical oversight, safe escalation, or independent auditability. Those claims should be verified rather than inferred from a companion-style interface.
• Engagement-loop design plus emotional intimacy creates a risk pattern. Yatri AI aims to reduce it through bounded reflection, tool confirmations, age-sensitive content limits, and routes back to human support.

Section 02

How the current Yatri AI controls differ

These are not policy promises. They are product constraints baked into the system.

• Yatri AI’s system boundary names it as a practical reflection companion, not a therapist, doctor, emergency service, or human friend. Dating, sexual, hookup, and adult flows are blocked for minors and people whose adult status is unknown.
• Crisis-risk language is classified before ordinary generation and returns a deterministic safety response with urgent-support signposting.
• Crisis-risk language is classified before ordinary generation. A crisis category returns a deterministic safety response and creates a safety-report path instead of asking an open-ended model to improvise emergency guidance.
• Yatri responses are designed to suggest a grounded next step outside the chat, such as journaling, breathing, preparing a conversation, or seeking appropriate human support.
• The public product should not be presented as universally suitable for minors; any age-specific surface requires separately verified enrolment, consent, and safety controls.

Section 03

How Yatri AI handles high-risk language and tool actions

Yatri AI’s most important property is not what it says — it is when it hands off.

• An ordinary reflection can end with a journal prompt, a drafted check-in, practitioner discovery, or another user-controlled next step.
• A save or side-effect action is proposed and waits for explicit confirmation.
• High-risk language takes a deterministic safety response and report path instead of an improvised clinical answer.
• The product states that urgent danger requires local emergency or crisis support and a trusted nearby person.
• These are documented Yatri AI boundaries, not a claim that another product lacks every comparable control.

Section 04

What good AI safety in emotional support actually looks like

These are the principles Soul Yatri uses to evaluate every Yatri AI feature before ship.

• Clear role boundary. The base safety prompt defines Yatri as a reflection companion rather than a therapist, emergency service, or human friend.
• Age-sensitive content limits. Dating, sexual, hookup, and adult flows are blocked for minors and users whose adult status is unknown.
• Deterministic high-risk handling. Crisis language takes a bounded safety response rather than ordinary generation.
• Deterministic crisis routing and a safety-report path for high-risk categories.
• Explicit confirmation before a private insight is saved or a side-effect tool action is executed.
• The response points back to real-world and urgent-support options without claiming to make a clinical judgement.
• The intended outcome is a user-controlled next step, not a promise of therapeutic benefit.