AI Companion Safety
Last updated: October 3, 2026
What It Does
Live AI Companion provides a short voice-based reflection space. While a session is active, microphone audio is streamed through Mind Shed's authenticated relay to OpenAI for a spoken response and transcription. Sessions are limited to 15 minutes. Saved summaries and transcripts remain on the user's device; Mind Shed does not intentionally persist raw audio or transcript content on its server.
Safety Protocol
- The trusted server configuration instructs the AI to avoid diagnosis and to direct crisis disclosures to immediate human support.
- Completed user transcript events are checked for self-harm or crisis language. Raw phrases are not written to application logs.
- When a safety signal is detected, the app surfaces crisis resources and the normal subscription gate does not block the safety response.
- Aggregate safety-event counts may be recorded without transcript text so the operator can monitor reliability.
- Users can end a session at any time and report harmful AI output to safety@mindshed.app.
Privacy And Control
Mind Shed requires explicit AI-processing consent before AI features are used. Consent can be withdrawn in Settings without deleting local history, planning, tasks, habits, or tracker data. See the Privacy Policy for data categories, processors, retention, and rights.
Scope And Review
Mind Shed is available only to adults age 18 and older. Safety logic and crisis routing are covered by automated tests and real-device release checks. This page describes the current protocol; material changes will be reflected here.