Your conversation.
How it’s handled.
Privacy policy, November 7, 2026. This page describes the HearMeOut iPhone app and browser conversation experience for adults 18 and older. The waitlist has separate information.
Your choice before AI processing
Before public-release features send content to AI services, we show the recipients and ask for explicit permission. Anthropic processes messages and transcripts for responses, moderation, coaching and recaps. Deepgram processes recordings for transcription and text for generated voices. OpenAI, through Vercel AI Gateway, processes avatar prompts, revisions and optional photos. Perplexity receives questions you explicitly ask to check for sources. If the language-model service is routed through OpenRouter, that service and the model provider are named in the permission screen before use.
You can decline or turn off AI sharing in Your space on iPhone or in your web privacy choices. Features that need this processing, including safety-checked human conversations, will not start while permission is off. Account access, reports and deletion remain available. Permission withdrawal stops future requests; it cannot recall requests already sent. A new AI recipient requires a new permission choice.
Your profile and permissions
We store your chosen display name, account identifier, selected avatar, adult confirmation and the versions and times of your community and AI permissions. We do not request your date of birth. Human conversations require an Apple-linked account. Apple verifies control of that account; we do not claim to verify your legal identity, age or beliefs. Your partner never receives your Apple account identifier or sign-in credentials. Sign-in tokens used for account recovery and revocation are protected on the server.
On supported iPhones, we ask Apple for an age range to enforce the adult-only experience. That system-provided range is used on your device; we do not send it, your birth date or any age-verification document to our servers or AI providers. The server keeps your adult confirmation. Where Apple requires age assurance, declining or an inconclusive result does not unlock the app.
What you share
Your display name and accepted voice notes are shared with the other participant, whether you meet through matching or a private invitation. HearMeOut’s server stores the conversation, audio, transcripts when available, timing information, and AI-generated coaching and recap. Conversations are not published to a public feed.
What AI receives
In matched conversations and new live invitations, recorded audio is sent to Deepgram for transcription, and the transcript is checked by the configured AI provider for safety before delivery. These checks are required for these conversations and can make mistakes. Some older invitations support optional cloud transcription. Transcripts may also be sent to the configured AI provider for a reflection check, claim classification, private coaching, and a shared recap. The provider may be Anthropic or a model accessed through OpenRouter. Your counterpart cannot see your private coaching.
Optional source research, where offered, sends the question you explicitly choose to research to Perplexity. It does not need your full conversation. Avoid putting names or private details into a research question. AI output can be incomplete or wrong; it is support for the conversation, not a judgment of either person.
AI practice
AI practice is a separate, simulated conversation. There is no human participant on the other side. The service stores your selected question, starting view, practice messages, recordings, generated replies, and any coaching or reports you request. Your messages and conversation history are sent to the configured AI provider to generate responses. Voice recordings are transcribed by Deepgram when you send them. Choosing voice mode, or explicitly requesting an audio reply, sends generated reply text to Deepgram to synthesize an AI voice. Playback is your choice, and practice audio can be replayed. Coaching is optional and requested separately after the practice ends. Source lookup sends only your explicit research question to Perplexity, not the practice history.
Unsent practice drafts stay on your device. A pending send keeps the same request identity until it is confirmed, so a retry does not silently become another message. You can delete a practice from its controls; this removes its content from the active service. Non-content cost and abuse counters may remain, and provider-held copies follow their own retention policies. AI practice is not published or sent to a human match.
Your optional clay avatar
When avatar creation is available, you can describe a look, upload one to three photos of yourself, or combine both. Your prompt, photo references, and an existing avatar when you refine it are sent through Vercel AI Gateway to OpenAI to generate your clay portrait. Photo references are optional. They are not shared with conversation participants or exposed through an image link.
Generated versions stay private until you choose one. Your selected avatar is visible to people you share a conversation with. You can change it or remove it in your clay studio. Reference photos are held in private storage for processing and scheduled for removal after the job ends or within one hour. Physical deletion can be delayed if cleanup fails. Provider-held copies follow that provider’s retention policy. Avatar jobs record their status and usage so that retries can recover the same request without silently starting another paid creation.
The browser saves your avatar draft and prepared photo copies locally so an interrupted request can recover. Successful creation clears the local photo copies; you can remove draft photos yourself before sending. Account deletion clears local drafts and prevents a delayed save from restoring them. Removing a saved avatar removes it from the active service and schedules its private storage object for deletion.
What stays in your browser
An HTTP-only session cookie keeps your signed-in profile available in the same browser. Browser recovery drafts, including audio and transcripts, are saved in IndexedDB on your device until a successful send or an explicit discard. Playback position, entry consent, and the reference to your conversation are stored in session storage to help you return to the same exchange and preserve the one-listen experience. Clearing browser data removes these local records and may end your guest session. A guest profile stays with that browser. When Apple sign-in is enabled, linking Apple lets you return to the same profile from supported devices. Separate profiles are never merged automatically.
A recovery draft is not yet shared with the other participant. Transcription, when enabled or required for matching, may already have sent its audio to Deepgram before you choose to send the note. Keep access to this device and browser private, especially on a shared computer.
The one-listen format
The interface gives the other participant one listening pass and a response window. This is a conversation format, not a guarantee that someone cannot make their own recording. Share only what you are comfortable sharing with the other participant.
Storage and retention
Vercel hosts the service, and Postgres stores account and conversation records. Voice audio is kept in Postgres or private Vercel Blob storage, depending on the service configuration; existing Postgres recordings remain there until migrated. Playback requires an authorized session. Providers may keep operational and security logs. Conversation records do not currently have automatic deletion after a fixed age. Review account actions in your browser account. Closing a browser page or discarding a local draft does not delete a voice note already sent to the server.
Deleting an account removes its profile and shared conversations from the active service for both participants. Cloud audio deletion is processed separately and can be delayed or require operator attention if retries fail. A later deletion pass checks for delayed uploads. Cleanup records keep an object reference and operational status, without your name, transcript, or conversation content. Backups and provider-held historical copies follow the providers’ retention policies; account deletion does not confirm immediate physical erasure of every copy.
Provider retry-result caches become eligible for cleanup after 24 hours, and matching outcome metadata after 30 days. Cleanup happens when another usage reservation is made, so deletion may occur later than those thresholds. Account deletion removes identity-linked data earlier.
To request access or deletion, use the support page or email team@tryhearmeout.com. We may need to verify your connection to the account or conversation. Do not include an invitation code or private transcript in a public message.
Matching
Before a human conversation begins, both people see an introduction and choose whether to accept. Matching stores your selected question, starting position, consent version, and queue timing to introduce you to someone with a different starting position. Your search has an expiry and checks whether you are still available. Blocking prevents future matching with that account. Choosing a side is a starting point for discussion, not a permanent public profile or a verified description of your beliefs.
Invitations and measurements
We keep account-linked usage counts, conversation timing and outcomes, requested source-search results, and technical diagnostics. These help enforce allowances, recover requests, handle failures and understand how well the service works. Operational reports can measure matching, conversation completion and provider performance. We do not sell this data or use it for advertising tracking.
An invitation link gives someone a way to request joining that conversation. Send it only to your intended participant. The site records waitlist referral parameters when you sign up. The browser demo choices stay on the page; they are not submitted as real opinions or conversations. The website does not currently use advertising pixels or third-party analytics.
Your choices
Older invitations may allow you to decline cloud transcription. Matching and new live invitations require transcription and safety checks; you can choose not to enter the queue. In either mode, you can pause before recording, review your own recording, and choose when to send. Microphone access is requested by your browser when you start recording. Remove that permission in your browser settings whenever you want.