Auralis Privacy Policy

Effective: November 2, 2026 App: Auralis (Auralis: Sound & Presence), iOS Developer: Aaron Monick ("we", "us") Contact: support@helixstudios.app

Auralis is an adaptive audio companion. You speak, music plays, and Auralis listens and responds when it would help. We designed it so that your words, sessions and journal stay on your iPhone.

The short version

What Auralis stores on your device

Stored locally on your iPhone, not on our servers (we have none):

iCloud sync (optional, off by default). If you turn on Sync with iCloud in Settings, your sessions, journal (including pictures, your photos and voice memos), the Library's written lessons and practices, and settings are also kept in your private iCloud database, so they match across your devices, and your AI keys sync through iCloud Keychain. This uses your Apple Account under Apple's terms; we can't see it. iCloud Keychain is end-to-end encrypted; your sessions and journal are too if you turn on Advanced Data Protection for iCloud. Turning sync off keeps everything on the device. With sync off, Auralis only checks whether its settings are already in your iCloud (from Auralis on another device, or from before it was installed again), to offer turning sync on; it takes nothing from there and sends nothing.

App lock (optional, off by default). You can lock Auralis with Face ID, Touch ID or your passcode in Settings → Privacy & Safety. The check is done by iOS; Auralis never sees your face or passcode. When the lock is on, the app is covered in the app switcher, your journal is removed from Spotlight, Siri won't read it, and the morning dream reminder doesn't show the intention you set for your dreams. A session is never interrupted by the lock, but if the app was away longer than a minute during it (an overnight Sleep session), it asks again when the session ends.

You can delete individual sessions (History) and journal entries (Journal), delete all journal entries (Settings → Data & Storage), or delete everything by deleting the app. A session or entry you delete moves to Recently Deleted (the Journal's More menu, or Settings → Data & Storage), where you can restore it or delete it right away; after 30 days it is erased. Like other apps, this data may be included in your own iCloud or computer backup of your iPhone, under Apple's terms.

Sharing an entry or letter. Sharing an entry or a Looking Back letter as an image or PDF makes that file on your device, from that entry or letter only, and hands it to the share sheet; where it goes is up to you.

Microphone and speech

When a session is running and listening is on, Auralis uses the microphone to transcribe your speech on your device, using Apple's speech recognition. Audio is processed in memory and discarded. It is never recorded, saved or sent anywhere. You can turn listening off at any time during a session.

Toning. In the Toning meditation, the same microphone audio is also analysed on your device to find the pitch you're singing, so Auralis can hold a drone at that note. Only the current note is kept, in memory, while the session runs; no audio is recorded or sent, and sung vowels are not sent to any AI model.

Automatic speech. In Journaling's Automatic speech, what you said is saved to your journal word for word (it syncs like any entry if iCloud sync is on), and at the end that transcript is sent once to the AI provider chosen for the session, to pick a few lines to read back.

Voice memos. Outside sessions, when you tap Voice memo in a journal entry, Auralis asks for the microphone (the first time) and records until you tap stop. The memo is saved with that entry only, as described above, and never sent to any AI model.

Song recognition

If you keep music from another app playing during a session (for example YouTube or Spotify), Auralis can recognize the song using Apple's ShazamKit. A few seconds of microphone audio are turned on your device into a signature (a fingerprint of the music that can't be turned back into sound or speech), and only that signature is sent to Apple's Shazam service to find the song. Auralis doesn't listen for songs while it is speaking. Recognition is handled by Apple under Apple's privacy policy.

Artificial intelligence

Auralis generates replies in one of these ways, depending on your settings:

  1. On-device (Apple Intelligence): on supported iPhones, replies are generated on your device. Nothing is sent anywhere. On devices without Apple Intelligence, you can choose in Settings to download a private open model (Qwen3 4B Instruct, Qwen3 4B or Qwen3 1.7B on a phone, about 1 to 2.3 GB, from Hugging Face). The download contacts Hugging Face's servers for the model files only; no personal data is sent. Once downloaded, replies are generated on your device and nothing is sent anywhere. You can remove the model in Settings. Natural voices: Auralis downloads an open voice model (Kokoro, about 90 MB, from Hugging Face) by itself on Wi-Fi when your device speaks English and has room, or you can download it in Settings → Voice. The download fetches the model files only; the voice is then made on your device and nothing you say or hear is sent anywhere. Unless you choose another voice, it speaks in place of Apple's voices, an Enhanced or Premium one you've downloaded too (with a subscription, Auralis Cloud's voice comes first): Auralis decides from the voices on your device and sends nothing to do so. You can remove it in Settings, and then it isn't downloaded again unless you ask. (From 22 to 25 October 2026 it also downloaded four British voices and their pronunciations where the device's region was the United Kingdom or a British English voice was installed; it no longer does, and deletes those files if they're there.) Hearing when you've finished (experimental, since 18 October 2026): if you turn it on in Settings → Voice, Auralis downloads Smart Turn, an open model (about 17 MB, from Hugging Face; the files only, no personal data is sent; kept out of your device's backups). On headphones only, while a session listens (never on the speaker, and never in Release), the last few seconds of the microphone's sound are held in memory for that model, on your device, to hear whether you've finished speaking. That sound is never stored or sent, and it's erased when the session stops listening or you turn the setting off. Your Mac: if you turn on sharing in Auralis on your Mac, your iPhone and iPad can use the Mac's larger downloaded model, at home or away. The devices connect directly to each other over an encrypted peer-to-peer connection (WebRTC); requests and replies go only between your devices. To set up the connection, each device writes its network addresses and a signed offer or answer to your private iCloud database, and asks public STUN servers (Google, Cloudflare) for its own public address; no conversation text passes through iCloud or any server. A shared secret in iCloud Keychain makes sure only your own devices can connect. On the Mac, a larger model can run in Auralis Model Host, a companion Mac app, which shares it with your iPhone and iPad the same way (your private iCloud for setup, the same iCloud Keychain secret, a direct encrypted connection); Auralis on the Mac can also pass it requests over the Mac's own loopback address, which nothing outside the Mac can reach. It does so only after Auralis Model Host proves it holds the same iCloud Keychain secret, and each request and reply is encrypted with it, so no other program on the Mac can read them. Auralis Model Host can also paint journal pictures, if you download its picture model from its menu (FLUX.2 [klein] 4B from Black Forest Labs and an image classifier that checks each picture, about 16 GB from Hugging Face; only the files are downloaded, nothing about you is sent, and it's never downloaded unless you say yes). A picture's scene description then goes only to your Mac, over the same direct encrypted connection (or the Mac's own loopback address, encrypted with the same secret), and the picture comes back the same way; your Mac keeps neither, and nothing leaves your devices to paint it. The picture model stays on the Mac (about 7 GB) until you remove it from the menu.
  2. Private Cloud Compute (since 25 October 2026): on devices with Apple Intelligence that offer it, Auralis uses Apple's Private Cloud Compute for a few background jobs, and for nothing else: a session's summary after it ends (the transcript, as much as Apple's on-device model would also read), the journal's tagging (each new entry's first 700 characters for its themes and whether it tells a dream, and the sentences around each new name to tell which names are people), and the occasional short question Auralis asks in a guided practice (a meditation, breathwork, sleep: what you've said in the session so far and Auralis's lines around it). Never a conversation's replies, never Need Help or the safety checks, never a voice command, and never while replies are Private or for a session that ran as Private: then Apple's model runs only on your device. It is on by default where it's offered; turn it off under Settings → AI & Models → "Allow Private Cloud Compute", and every job stays on your device. Apple designed Private Cloud Compute so your data is not retained and is not accessible to Apple. It is asked only where Apple's model would have been (when Auralis Cloud, your own key or your Mac does the job, it isn't), and if it can't answer (offline, Apple's daily limit, an error), the job is done on your device instead.
  3. A cloud AI service with your own key (optional): Auralis asks you first, once for each provider on a device, as you save its key or connect your server: what goes there, and that its own terms apply; nothing goes to it unless you allow it (since 22 October 2026). If you add your own API key for a provider (Anthropic, OpenAI, Google Gemini, or a server you set up), the text of the current conversation, and, in Journaling, Custom, Record a Dream, Dreamwork and Active Imagination sessions, a few passages from your past sessions or journal that are relevant to what you just said, is sent to that provider to generate a reply. In Active Imagination's Light and Structured levels only what you said since Auralis last spoke (or in the step you're in) goes, to word a short prompt or the next step's opening, and nothing from your past; in Witness nothing is sent while you speak (since 25 October 2026). Now and then a Journaling, Custom, Inquiry or Dreamwork session opens by picking up something from your last two weeks (Settings → Journal → "Pick up where we left off", on unless you turn it off): the one sentence you wrote or said about it (never from a session that ran as Private), or a dream image, goes to that provider to word the opening line (with no key, your Mac's model words it, or a set line is used and nothing is sent). Which ones were picked up is kept on this device. If you've added keys for more than one and the one you chose is unavailable, Auralis may use another one you added. If you've turned on any traditions (Traditions, at the bottom of Practice), their names and a sentence about each go with the request (replies, Looking Back, the briefing for readings), so they can use their terms. In Conversation, Dreamwork, Active Imagination, Learning and a tradition's practices, a few short passages from the Library's own courses (Auralis's text, chosen on your device by matching what you said) go with the request too; nothing of yours is added to them. A Conversation (since 30 October 2026; "Talk with Auralis", a kind of the Journal door since 31 October 2026) sends only the text of the conversation and those Library passages, never passages from your journal or past sessions. Every Journal-door session (Talk with Auralis, Reflect, Work through something, Inquiry) ends with one choice before anything is written: "Save to journal" (the default) or "Let it go". Saving writes one ordinary journal entry (a title and one-sentence summary, then the whole talk with "You:" and "Auralis:", your words as recognised, without voice commands or Auralis's fixed safety lines); it follows the journal's rules (on your device, in your iCloud if sync is on) and is sent nowhere; the only request it makes is the one for the title and summary, to the model that session used (nothing is sent if there was none). "Let it go" erases the session, its transcript and its events from your device: no entry, no summary, no history. What a model heard while the session ran was already sent, and Auralis says so under the button. Leaving without choosing saves it, as does reopening the app after it was closed mid-session. Talk with Auralis can also be a talk with a person (since 31 October 2026): a thinker from the list, whose own words the Library holds, or anyone whose name you type. Such a talk needs a larger model (Auralis Cloud, your own AI key or your Mac's model; never Apple's or this device's own model). Its requests carry the conversation, a few short passages of that person's own words from the Library (when it has some) and notes about them from its courses, and that person's memory of you (below); nothing from your journal, your portrait, your past sessions or any other person's memory. The names you type are kept on this device for quick reuse, never sent anywhere except as part of that talk's requests. A person's memory of you: after a talk with someone that you saved, the model that session used is sent that person's memory so far and the talk, and writes it again: a short note (under 250 words) of what you talked about, what you shared and what they suggested. It is kept on this device only (not with your sessions, and not in iCloud), sent only with that same person's talks, and shown in Settings → Journal → "What … remembers", with Forget. A talk you let go is never remembered; with no model, nothing changes; Delete All in the journal forgets every person's memory too. Tradition reminders are scheduled on your device; nothing about them is sent anywhere. When you ask a question about your past sessions or journal (in Search, with Ask Auralis on Today, the Journal, Siri, Control Center or the widget, or by voice in a session), the question and the passages that match it are sent to check the matches and write the answer; nothing else from your journal goes with them. The questions offered under "Ask about your journal" are written by the same AI that answers it (your own devices when replies are Private), from short excerpts of your last entries and the themes, dream images and names Patterns found, sent when the questions are first shown after Auralis opens, then again only on a new day or when your recent entries change. A question about sleep, heart or rest is answered with Apple Health too: when your own Mac's model or a model on your device writes the answer, it's given your last four weeks of nightly sleep hours, resting heart rate and heart rate variability, and the patterns Auralis found. A cloud AI service is never given those numbers; it gets only the pattern sentences, and only if you turn on Include health patterns with cloud AI. Siri and Shortcuts open the answer only on an unlocked device, and the app's own Face ID lock still applies. When you ask Practice to make a path from your journal, the theme you chose and up to six entries about it are sent to write the days. The title and artist of the song playing are included so replies can refer to it, and the local time of day, so a question about the time can be answered. In the Journal, features you start yourself send the entries they work on: reflecting on an entry sends that entry (with what you answered in Ways of seeing it about its images, unless it came from a session that ran as Private) and up to three earlier ones; Ways of seeing it, on a dream or an inner practice's entry, sends that entry, what you answer about its images, up to two earlier passages where each image (or a person in it) appeared and what you answered about it before, and any meaning you wrote for it in the Symbol Library, only to Auralis Cloud, your own key's provider or your Mac (a Private entry, or with Settings on Private, only to your Mac); a symbol's thread sends the entries that mention it; Looking Back sends every entry in the span you chose (a week, a month or three months), and, if Write the weekly letter by itself is on (Settings → Reminders; on if you've had a letter written before), the past week's entries once a week when there are three or more, only to Auralis Cloud, your own key's provider or your Mac; adding a session to your journal (and, when it ends, a Record a Dream session, unless you turned journal entries after sessions off) sends what you said in it (or, if you said little, its notes) to be made into an entry in your own words; proofreading an entry sends that entry. Without being asked, Auralis also tags your journal: it sends each entry's opening lines (its first 320 characters) to tell dreams from other entries, and its first 700 characters to find its themes, once (again if you edit it); and each new name it finds, with up to three sentences it appears in (each up to 160 characters, from your entries or a session's intention, summary, marked moments or what it worked with), to sort which names are people, pets, public figures or you. An update that changes how Auralis tags may send them again. This goes to your provider or Auralis Cloud only while you're online and replies aren't Private; offline, with Private, or with no key or Auralis Cloud in use, your own devices' models (your Mac's, Apple's or the one you downloaded) do it, and nothing is sent. When enough new words recur, the list of your journal's recurring words (single words and names, no sentences) is sent to group the words for one thing (snake and serpent): only to your provider's strongest model or Auralis Cloud's while you're online and replies aren't Private, otherwise your Mac's large model, never a model on this device. That data is handled under the provider's own terms and privacy policy. In Storytelling and Learning, the topic you give is sent too; if you choose to continue a dream as a story, that dream's text is sent as the topic. In Dreamwork, the dream you choose is sent so the replies can work with it. In a Shamanic Journey told aloud with Light guidance or the Journey guide, what you say during the drumming (and the question you bring) is sent for its short lines only to your provider's strongest model, Auralis Cloud's or your Mac's large model, never a model on this device; with Speak freely, or with no such model, nothing of it is sent (the written guide runs on your device). Auralis never sends audio.
  4. Built-in responses: without any of the above, Auralis uses fixed, on-device responses and sends nothing.

Which one answers. There is one choice, in Settings → AI & Models, and every session follows it, however it starts (its setup, Start again, Today's cards, Your week, a path, a flow, a widget, the Watch or Siri). Sessions no longer have an AI choice of their own (removed 26 October 2026); a session's record still notes which choice it ran with. With Private, nothing goes to a service: only the models on your own devices (1: your Mac's, Apple's on-device model, or the one you downloaded) answer, a voice on your device speaks and reads aloud, and a picture's scene is described on your device (only that description goes out to be painted, when you ask for a picture in one of the services' styles; in one of Image Playground's own, it's painted on your device and nothing goes out; in ChatGPT's, only its short phrases go, through Apple, as described under journal images below). With Automatic or Cloud, what a reply or a feature needs goes to the service in use, as this section describes: Auralis Cloud's relay, which passes it to Together AI's open models (Cloudflare Workers AI as the backup) and doesn't log or store what you say (see Auralis Cloud below); your own key's provider, under its own terms; or, for a few background jobs, Apple's Private Cloud Compute (2). What a session that ran as Private said (set to Private in its setup before 26 October 2026, or run with Settings on Private) stays out of what goes to a service later, as described above. A journal entry you keep from a session is part of your journal like any other entry: Ask, Reflect, Looking Back and the journal's threads use it with the AI you chose in Settings. With Cloud, only the cloud services answer, your provider (3) or Auralis Cloud (offline there are built-in responses). With Automatic, Auralis uses the best available: your provider (3) or Auralis Cloud first, if you've added a key or subscribe and are online, then your Mac's model, then Apple's on-device model, then this device's downloaded model, then built-in responses. So with Automatic and a key, conversations are sent to your provider whenever you're online; choose Private to keep them on your devices.

Replies started early. With your own key, Auralis Cloud or your Mac's model, a reply may be started a moment before your turn closes, once what you've said sounds finished (since 18 October 2026): the same conversation is sent, with your words so far. If you go on, or something else should come first (a command, a safety check-in), it is cancelled, and nothing of it is said, shown or kept; with Auralis Cloud it still counts toward the day's requests, so a session stops starting them once most go unused. A model on your device is never asked early.

Safety check. When something you say in a session sounds heavy (a few words checked on your device first), the same model that answers is also asked whether it might be a wish not to exist: it gets what you just said (from a long stretch of talk, only its heavier sentences and the one before each) and the two lines before it, nothing else. Its one-word answer can bring up Need Help; it is kept only while the session runs and is never saved or included in exports or problem reports.

Auralis Cloud (optional subscription). If you subscribe to Auralis Cloud and it is the service in use (replies Automatic or Cloud), the request a reply needs — the same text an external provider would receive — goes over HTTPS to Auralis's relay, a small program on Cloudflare Workers, which passes it to the company that runs the open model (now Together AI, running DeepSeek V4.1 Flash for replies, and a larger open model such as DeepSeek V4 Pro for reading across your whole journal, up to 30 times a day; named in Settings → AI & Models → Auralis Cloud) and streams the answer back. If that company's service is down or too slow, the relay asks a backup instead: an open model run by Cloudflare Workers AI on Cloudflare's network (the same company that runs the relay), which receives the same request (shortened to fit that model if it's very long) and keeps neither it nor the answer. With the request, the app sends the App Store's signed record of your subscription; the relay asks Apple whether it is active and never learns your name or Apple Account. The relay does not log or store what you say or what the model answers, nor the model service's error messages: it keeps only a flag saying the subscription is active (for up to a day, checked with Apple again after an hour), counts of today's use (requests, journal reads, pictures, characters spoken; for up to two days) and the month's estimated cost of your use and a mark that the subscription was used that month (until the month is over), all filed under a one-way hash of the subscription's id. The month's figure is how Auralis Cloud keeps each subscription within a monthly allowance: past most of it, journal reads use the everyday model and pictures pause; past all of it, Auralis Cloud pauses until the 1st. It also keeps anonymous daily totals for running the service (how many replies, voice lines, pictures and journal reads were asked for that day, how many failed and with which status, and which model answered, and what that used: tokens, characters and pictures, and their estimated cost; how long replies took to their first words and to their end, by model and by feature, and how long voice lines took to start, in a few time ranges; how many subscriptions came near a daily limit or the monthly allowance; how many subscriptions used it that day and that month; and, per feature, how many requests and their estimated cost). Each request the app sends to Auralis Cloud names the feature that sent it, as one fixed word from a list (for example "lookingBack", "journalTagging" or "voice"), so these totals can say what the requests were for; the word is counted with the day's anonymous totals and kept nowhere else. These totals have nothing in them about who asked or what was said; they're kept up to 120 days. For the same reason the relay keeps its last 50 failures: when, which kind of request (a reply, a voice line, a picture, Apple's check), the status, the model and a short error code; nothing about who asked or what was said, and not the services' error messages. To spot unusually heavy use, a subscription's estimated cost for the day and the month is kept with its other counts under the same hash, and a hash whose day or month passes another 25 cents is listed with that figure (the day's for eight days, the month's for forty); nothing else about you goes with it. The developer sees these totals and lists on a private page only they can open. From that page the developer can pause Auralis Cloud for one subscription (for example, one being misused): only its hash and the date are kept, its requests are then refused before anything is sent on (the app says so and your other models answer), and the day's totals count pauses and refused requests as numbers. That page also shows App Store Connect's own aggregate reports, which the relay fetches once a day with the developer's App Store Connect key: downloads, active subscriptions by product, subscription events, installs and deletions, devices, sessions and crashes (these last from people who chose to share analytics with app developers, and only where at least five people are counted), and the number of TestFlight testers and each recent build's installs, sessions and crashes. Only the daily numbers are kept (for 35 days); Apple's reports name no one, and nothing in them is linked to your subscription's hash or anything you said. When something fails or slows down, or a build of the app crashes more than before, the developer gets an alert with counts, times, status codes, model names and app versions only (for a crash, also its kind and its place in the app's code, from the anonymous performance reports below). Apple also tells the relay when a subscription is refunded, renewed or lapses (its signed notice is checked as the app's is), so the "active" flag is updated at once; only the notice's type is logged. The developer is told that someone subscribed or a subscription ended, and gets a short morning summary of the day before in counts (requests, failures, estimated cost, how many subscriptions used it, subscriptions started and ended, App Store downloads and active subscriptions, crashes and hangs by app version, how many anonymous reports came in, the most used features and how many finished or skipped the first-launch pages), never who. The model service is used under a contract term that it keeps no prompts or answers and trains on none. If your sessions use an Auralis Cloud voice (with a subscription they do, unless you choose a voice on your device in Settings → Voice), each line Auralis is about to say is sent the same way, through the relay to the same company's open voice model (Kokoro), with the voice you chose (one of four: Heart, Bella, Nicole or Michael; if that voice is down, Cloudflare Workers AI's voice model speaks the line instead, on Cloudflare's network); the audio comes back and is kept only in this device's cache. When the voice you chose is one of the natural voices on your device (all four are, once they're downloaded, in English), a line is made on your device instead and isn't sent, with the screen on or locked: a line is sent only before the voice on your device is ready, or for the rest of a line it couldn't finish. Reading aloud works the same way, except in Listen in Auralis (the share sheet): iOS doesn't give it room for the voice on your device, so with a subscription it reads what you shared in the same voice from Auralis Cloud, sent the same way. The relay keeps a count of characters spoken today, under the same hash, and nothing else. Cancelling the subscription stops all of this; Private keeps everything on your devices, voices and reading aloud included.

Paths that remember. A day of a path builds on the days before: the short summaries of the sessions that completed its earlier days go with that day's session to the model that answers, and to the same model once while you set it up, to write a line linking them to today. Nothing else from those sessions is sent.

The journal's briefing While "Journal informs conversations" is on (Settings → Journal), what goes with a request includes a few lines from each of your last several journal entries and recent dreams (from the past three weeks), your journal's recurring themes this month, your last session's summary and your path day, and a private portrait of your life: a briefing on the experiences, people, places, ideas and dreams in your journal, which a model writes from your whole journal and then updates with your new entries. The portrait is kept on your device (not with your sessions, and not in iCloud), and deleted with your journal; a line of it resting on an entry you've deleted isn't used. Writing or updating it sends your journal (or its new entries) to the model doing the writing: your AI provider's strongest model if you've added a key and chose Automatic or Cloud, otherwise your Mac's model. It is never written by a model on this device, which isn't accurate enough for it. A second pass by the same model checks the portrait against your journal, line by line, and removes anything the journal doesn't support (so your journal goes to that model twice). Each line quotes the words of the entry it comes from; your device then drops any line whose words aren't found in your journal (this check stays on the device). The same goes for Looking Back's letters, the threads of a symbol or a dream series, reading your dreams for series (and checking the series found by shared images, and "Add a series"), and writing a personal path. Without a key or your Mac, those wait until one is available. Reading your dreams for series, and checking those series, happen without being asked (once when you have enough dreams, and when you open your series): every dream's text goes to that model, never to a model on this device. Opening a series found by shared images that has no name yet sends the start of each of its dreams (up to 500 characters each) and the shared images to name it: to your key's provider's or Auralis Cloud's everyday model while you're online and replies aren't Private, otherwise to your own devices' models.

Speak freely (Release, a kind of the Journal door, and "Let this one burn" for a new written entry) keeps only that it happened unless you choose otherwise: its kind, when it started and ended, its length, your arriving and leaving check-ins, and a timeline without any text (when the bells rang and the page burned). At the closing bell you choose (since 31 October 2026): "Let it go" burns your words and nothing of them was ever kept; "Save to journal" writes them, exactly as recognised, as one journal entry (on your device, in your iCloud if sync is on, sent nowhere, no model asked to read or summarise it). Until you choose, never your words or Auralis's, never a journal entry (whatever the journal is set to), nothing for Spotlight, Siri, the Live Activity or the Watch beyond its name, and nothing that could bring the words back if the app closes during it: closing the app before you choose keeps nothing. A flow's Release part still burns at once. No AI model hears them; the one exception is safety's second look at something heavy the phrases miss, and only on a model on your device (Apple's own, or one you downloaded), never a cloud service or your Mac. The words are held in memory until they burn. A new written entry you let burn was kept as a draft on your device while you wrote it, as any entry is; the draft is deleted the moment you let it burn.

The week in review (a kind of journaling) tells the model what it tells you at the start: how many sessions you finished in the last seven days and their names (the practice, or a flow's name). No dates, times or what was said in them. The day in review does the same for today: how many sessions you finished today and their names. A recurring feeling sends the feeling you typed, if any, and Something on your mind what you typed about it, if anything, with your words, as journaling does.

Planning Your week for you sends the same model the names of the practices you did in the last month and how often, your journal's recurring themes as single words (not your entries), the traditions you've turned on, what you type about the week, which practices your own check-ins show usually leave you calmer or with more energy (their names and how often, never the check-ins themselves), and, when you've done one in the last week, the summary of your week in review (not one that ran as Private, when the week may be planned by a service). Which practices help you is worked out on your device from your check-ins and isn't kept or sent anywhere else. Nothing is sent until you ask for a week, and a balanced week needs no model at all.

The Library (an optional tab) keeps its courses and your progress on your device, synced through iCloud with your settings when sync is on. Since 1 November 2026 the lessons written out for you, and the practices written from a course, sync with your journal instead (in your private iCloud only, with sync on), so another device of yours reads them rather than having them written again; a practice you remove is removed on each of them. The built-in courses' lessons and their recall questions come with the app: reading or listening to one sends nothing anywhere. When you add a topic, what you typed and the names of the traditions you've turned on go to the same model (the strongest hosted model with your key, else your Mac's) to write its course. Going deeper into a finished course, writing a tradition's course afresh, or writing out a part that didn't come with the app (a part of a topic you added, a branch, Go deeper's new parts, or a built-in part changed since its lesson was written) sends the course's outline (not your journal) to that model, and fact-checking a written course sends its outline to the same model. A practice from a course (since 28 October 2026: Make a practice from this course, or from a part) sends that course's or part's outline and the length you chose (not your journal) to the same model to write it; the practice is kept with the course, on your device and your others as above, and a session's setup made from it (a favorite, a day of your week, a flow) carries it as other settings do. "Suggest more for me", and "Generate more" on a shelf's page, send your traditions and the themes of your sessions and journal (single words, not your entries), with the names of the Library's own courses and earlier suggestions so that what comes back is new, never the name of a topic you added; they and the recall questions for a part that didn't come with its own (which send that part's text, not your journal) go to Auralis Cloud's or your key's everyday model while online, unless replies are Private, otherwise to your own devices' models. A Library story is told in a Storytelling session like any other; a course's lessons are read, or read aloud as any text is. A teaching flow (since 2 November 2026: on Flows, a talk read from a course you have, a sit, then questions) reads the part's lesson aloud in a session as read aloud does. When the next part's lesson isn't written yet, it's written ahead by the same larger model that writes lessons (its outline, not your journal), and a question to reflect on is written ahead from the part's outline (nothing of yours); both are kept with the course as above. Questions you ask in its last part are answered from the talk's own text, sent with your words to the model your "Where replies come from" setting chooses, as any session's reply is (open questions only with a larger model: Auralis Cloud, your key or your Mac). Where the series has got to, last time's point, and each question you asked there with the first sentence of its answer (never the whole session; none from a Private session) are kept with your Library progress, on your device and synced through iCloud as it is, so the next talk goes on where the last stopped and can follow up your question. Asking about a lesson sends your question and that part's text (not your journal) to the model your "Where replies come from" setting chooses, as a session's reply does. "Where this shows up for you" sends a few journal entries that touch the part (only those) to the model your "Where replies come from" setting chooses: with Automatic, your AI provider (3) if you've added a key and are online, else your Mac's model or this device's; with Private, only your own devices. The Library's own courses come with their icons and their feed's pictures, painted once for everyone before the app was built, so nothing is sent for them. A topic or path of your own (or a branch you follow) without an icon gets one drawn by OpenAI's image model when you've added an OpenAI key, or else by your Mac's picture model when it's connected, or else (in TestFlight builds only, for now) by Auralis Cloud's open image model when you subscribe: its title and a line about it are sent, nothing else; without any of those the icon is a built-in symbol. The Library's feed is put together on your device. Posts written for it are written from the courses' text (never your journal) by the model your replies come from (with Automatic, your hosted provider and key while online, then your Mac's model, then this device's; with Private, only your own devices), and kept on your device. What you linger on, save, open and mark "More" or "Less like this" is kept only on this device to order the feed, and is never sent anywhere. The feed can also show your own practice from this device: a session from about a week ago (what it was, how long, its summary's line), a path's next day, passages you kept in the Library with your notes, and open questions Looking Back drew from your journal; these are read on the device to put the feed together and are not sent anywhere. With an OpenAI key, your Mac's picture model or (in TestFlight builds only, for now) an Auralis Cloud subscription and "Pictures drawn for posts" on, a few posts a day from courses of your own and branches you follow get a picture drawn (by OpenAI's image model with your key, else by your Mac's, else by Auralis Cloud's open one): only that post's short title and first words are sent (never a recall question, never anything from your journal), and the picture is kept on your device. "A daily lesson" is a notification scheduled on your device with a line from a course you follow; nothing is sent anywhere. Sharing a topic makes a link you send yourself: a built-in topic's link carries only its name, and one written for you carries its course outline, glossary, texts and stories (not your progress, your journal or the lessons written out for you).

The same portrait (shortened for a model on your own devices) also goes with each request in conversation sessions (Journaling, Dream sessions, Active Imagination, Inquiry and Custom), so Auralis understands who you are; with a key, starting one of those sessions may update it in the background. Turn "Journal informs conversations" off to keep all of this out.

OpenAI voices and journal images (optional)

So there's no silence between sentences, the next sentence of a reply is sent to the voice service you chose (OpenAI or Auralis Cloud) while the one before it is being said. If you speak over Auralis or stop it, that sentence may have been sent and never spoken. In the same way, a guided practice's next one or two scripted lines are sent before they're due, so they play without waiting on the network; a line with your own words in it (an intention, a dream sign) is sent only when it's said. If the session ends first, those lines may have been sent and never spoken. With an Auralis Cloud voice that's also on your device, these are sent only while lines would be said from Auralis Cloud (the voice on your device not ready yet); if it stops being ready, the next sentence and the practice's next lines are sent at once.

If you add an OpenAI API key and choose an OpenAI voice (and replies aren't Private), the text Auralis is about to speak is sent to OpenAI to turn it into speech: Auralis's own replies and guidance, and any text you ask it to read aloud, such as a journal entry or text you share to Auralis from another app. Your voice is never sent.

Pictures for journal entries are offered only with your own OpenAI key, your Mac's picture model, or (in TestFlight builds only, for now, since 25 October 2026) an Auralis Cloud subscription; the App Store app with Auralis Cloud alone, another provider's key or no key doesn't offer them. If you ask Auralis to make an image for a journal entry, the entry's text is sent to your AI provider (above) to describe a scene (with Private, a model on your own devices describes it, so the entry doesn't leave them), and only that scene description is sent to paint the picture: to OpenAI with your key; else, when your Mac is connected and Auralis Model Host there has its picture model, to your own Mac, which paints it there (see Your Mac above: the scene and the picture go only between your own devices); or, without either (or when they can't), in TestFlight builds, to Auralis Cloud, whose relay passes it to an image model run by Cloudflare Workers AI on Cloudflare's network, the same company that runs the relay (Leonardo's Lucid Origin or Phoenix, or the open FLUX.1 [schnell]; Together AI only if those fail) and keeps neither the description nor the picture; a subscriber can make up to 20 pictures a day. The image is stored with the entry on your device.

Both are handled under OpenAI's terms and privacy policy.

On a device with Apple Intelligence you can also choose one of Image Playground's styles (Illustration, Sketch, Animation). Apple's Image Playground then paints the picture on your device from the scene description, and neither the description nor the picture is sent anywhere to be painted: its sheet opens with the scene, and you keep the picture or close it. The scene is still described as above, and the same model turns it into a few short phrases for Image Playground: with Automatic or Cloud by your AI provider (the entry's text, then the scene, go to it, as for any picture), with Private on your own devices, so nothing leaves them. With Private, your Mac's four styles come first in the picture menu, painted on your Mac and never handed to a service (when it can't, Image Playground paints it on your device); a service's style says it sends the scene. When none of OpenAI, your Mac or Auralis Cloud can paint a picture you asked for in one of the four styles (no key, out of credits, today's pictures used, paused or failing), Image Playground on your device paints it instead, in the nearest of its own styles (never ChatGPT's). ChatGPT's styles in Image Playground (oil painting, watercolor and others, chosen in its sheet) are a separate choice, last in the picture menu under "Sends the phrases to ChatGPT, through Apple", and never made for you: only when you pick it does Image Playground send the picture's short phrases (the scene itself if none could be written), and anything you add in its sheet, to OpenAI's ChatGPT through Apple's ChatGPT integration, under Apple's and OpenAI's terms, as Apple's ChatGPT setting describes.

Beta feedback (TestFlight only)

In test builds, Auralis may ask after a session how it went. If you choose to send feedback, it goes by email from your own Mail app (or the share sheet), and you see it before it's sent. It contains your note, the app build and device model, the kind of session and when, and, only if you leave Include the transcript and logs on, that session's setup (never your trusted contact or the life portrait), transcript, timeline and technical logs. We use it only to fix problems, and delete it on request. Your answer (worked well, or something was off) is also kept with the session, on your device and in your iCloud like the session itself, never your note. The App Store version doesn't ask.

User tests (TestFlight only)

In test builds, Settings → About → Run a User Test records Auralis's own screen and the microphone, only after you read what's recorded and tap Start the Test, and only until you finish, tap Stop, or leave Auralis for more than a minute. Nothing outside Auralis is recorded. Beside the recording is a small file of markers: the app's build and version, the device model, the iOS version, when the test started, and when each task began and ended (no words and nothing from the screen). Both stay on your device, left out of backups, and are never sent anywhere by Auralis; you choose whether to share them with the share sheet. You can delete a test at the end or later in Settings; a test you haven't shared is deleted after 30 days. The App Store version doesn't have it.

Reporting a problem

Settings → About → Contact Support, and Report a Problem in each session's review, let you email the developer a question, a problem or feedback. It goes from your own Mail app (or as a file you share), and nothing is sent until you tap Send. It contains what you write, an email address if you add one, the app version and your device model. Only if you turn on Include diagnostics (offered for a problem), which you can read in full first, it also contains the AI setting and the models that answered in that session, the session's timeline of events (their kinds, timings, states and model names, never what you said, Auralis's replies or your journal), and the app's audio, voice, sync and Live Activity logs from around it, which record states and counts only, and the reports iOS makes for the app about once a day (MetricKit: hangs, crashes, slow launches and where in the app's code they happened; kept in full only on your device until you include them). We use it only to fix the problem, and delete it on request.

Performance reports (on unless you turn them off)

About once a day iOS gives Auralis a report of how it ran on your device (MetricKit). With Settings → Privacy & Safety → Share anonymous reports on, which it is unless you turn it off (and only once Auralis has told you, below), Auralis reduces each report on your device to a small summary and sends it to Auralis Cloud's relay, whether or not you subscribe. The summary holds only: the app's version and build, the iOS version (like 27.0), the device model (like iPhone17,1), whether it's a TestFlight or App Store build, and numbers: how long launches and resumes took and how long hangs lasted (in a few ranges), CPU and GPU time, time on screen and in the background, peak memory, disk writes, data used over Wi-Fi and cellular, location use, and how many times the app ended and why (closed normally, ran out of memory, crashed, too slow). For a crash or a hang it holds the kind of problem (the exception and signal) and one place in Auralis's own code (a position in the app's binary, like "Auralis+0x1a2b3c"), nothing else of the report: no other code, no file paths or names, nothing you said or wrote, no journal, no account, subscription or device identifier. It's sent without your subscription's token. Development builds never send.

The relay checks every field, refuses anything else, and doesn't keep the report: it adds it into daily totals by app version (and, for launch time and memory, by device model), which only the developer sees and which are deleted after 35 days. Like any connection, the request carries your network address; the relay uses it only to limit how often one address can send (at most 10 a minute) and doesn't keep it. A report that can't be sent waits on your device and is tried again when you next open Auralis; turning the switch off deletes any waiting and sends nothing more. The full reports stay on your device, as before.

Telling you first

Share anonymous reports is on unless you turn it off. Auralis tells you so when you first open it, in a line on its privacy page; if you had Auralis before, a small notice on Today tells you once instead, with OK and Turn off. Until you've been told, nothing is counted or sent: not the reports, and not the counts below. Once you have, the counts include which first-launch pages you'd seen, as described below. If you said no before (Don't share when Auralis asked, or the switch turned off), it stays off: Auralis never turns it back on for you. You can turn it off, or back on, any time in Settings → Privacy & Safety; with iCloud sync on, the setting, and that you've been told, are the same on your other devices.

Why we send them. We send these reports on the basis of our legitimate interest in keeping Auralis working and improving it: finding crashes, hangs and the problems people hit, and seeing which features are used. They hold no words and nothing that identifies you, and you can turn them off at any time; turning them off stops them at once and deletes anything waiting to be sent.

Usage counts (on unless you turn them off)

The same switch, Settings → Privacy & Safety → Share anonymous reports, also sends a day's counts of what was used in the app, to help make Auralis better. During the day Auralis counts on your device how many times certain things happened, from a fixed list, and nothing else: sessions started (by kind, like Meditation or Sleep), how long they ran (in a few ranges: under 5 minutes, 5–15, 15–30, 30–60, an hour or more) and whether they ended early; journal entries saved (typed, spoken, or from a session) and dreams recorded; Library lessons opened and finished; path days finished; Looking Back letters; Ask questions asked; journal pictures made; on the first-launch pages, which pages were reached, where the rest was skipped, and the choices made there (how to begin, which kind of first session Begin opened, like a breath sit or a dream told, Automatic or Private replies, and, when a tradition is what brought you, how many traditions were picked on its page: none, one, or two or more, never which; versions before October 21, 2026 also counted whether the Library was left on), and that you turned these reports back on in Settings after turning them off (the versions that asked first, in October 2026, counted where a yes was given instead); problems (the cloud voice handing a line to another voice and why, in a few kinds; a session's problems, like music that wouldn't start; replies a model couldn't give, by the kind of model; how many times a safety check-in was asked or Need Help opened, a number only); and on Auralis Cloud's page, whether it was opened (and from where: Settings, the first-launch pages, or a screen offering it because nothing on your devices could do that work, along with how often such an offer was shown) and a purchase started, finished, cancelled, failed or restored.

The counts never hold anything you said or wrote, the name of anything you made (an entry, a dream, a topic, a path), your traditions' practices, a time finer than the day, or anything that identifies you or your device. The day after, the next time you open Auralis or leave it, the day's counts go to Auralis Cloud's relay with the app's version and build, the iOS major version (like 27) and whether it's a TestFlight or App Store build, without your subscription's token and without your device model, whether or not you subscribe. Development builds never send.

The relay accepts only the names on its own copy of the list, limits every count, and doesn't keep the report: it adds the counts into daily totals for everyone together (the first-launch counts also by app version), which only the developer sees and which are deleted after 35 days. As with performance reports, your network address is used only to limit how often one address can send and isn't kept. Turning the switch off deletes the day's counts and anything waiting to be sent, and nothing more is counted.

Auralis Cloud's page also asks the relay how much of today's allowance you've used. That request carries only your subscription's signed transaction; the relay answers from the counts it already keeps for its daily limits and stores nothing new.

Sharing a summary (optional)

Journal → ⋯ → Share a summary makes a summary for a clinician, therapist or guide: the sessions and entries you choose, as a PDF and a plain-text copy. It's made on your device and goes only where you send it through the share sheet; we never receive it. Sessions appear as facts (when, what, how long, your check-ins); entries in your own words. Your saved AI reflections and your health patterns are included only if you turn them on. With the app lock on, Face ID (or your passcode) is asked first.

Reminders that follow your rhythms (optional, off by default)

A bedtime reminder at the time you usually start Sleep (worked out on your device from your Sleep sessions), and a gentle reminder when a path or course has been quiet for three days. They are notifications scheduled on your device; nothing is sent anywhere. A path made from your journal is never named in a reminder.

Stories from your own imagery (optional)

In Storytelling, "Draw on my journal and dreams" gives the storyteller up to six single words that recur in your journal and dreams (such as "river" or "owl"), found on your device. Only those words go to the model telling the story (the one your AI setting chooses), never what you wrote, and the names of people only if you turn that on.

Apple Health (optional, off by default)

If you turn on Log to Apple Health in Settings (or say yes when the review of your first Meditation, Breathwork, Shamanic Journey, Through an album or Explore a question session offers it, which logs that session too), Auralis writes these entries to the Health app on your iPhone: Mindful Minutes for finished Meditation, Breathwork, Shamanic Journey, Through an album and Explore a question sessions; a workout for each finished Movement session (its kind, such as walking, yoga or tai chi, and its start and end time, nothing else: no route, heart rate or calories; Apple's Health sheet asks for this as your first Movement session starts), or, when your Apple Watch with Auralis records it, the Watch's workout instead (see Apple Watch below), never both; and State of Mind from your arriving and leaving check-ins.

Separately, only if you turn on Show sleep and heart patterns in Settings (or say yes when the review of your first Sleep session offers it, if you're new to Auralis), Auralis reads your sleep, resting heart rate, heart rate variability and the mindful minutes other apps logged, and works out on your iPhone how they go with your sessions, your check-ins, your journal's themes and the dreams you record (for example, how you slept on nights after an evening session). The patterns are shown on Today and in Noticing. Your Health data itself is never sent to a cloud AI service, to iCloud or to us; the only model that ever sees its numbers is your own Mac's (over Auralis's link) or one on your device, when you ask about sleep or heart.

A Looking Back letter reads only your journal: it's never given these patterns or anything else from Health. You can revoke or change any of this at any time in the Health app or in Auralis's settings.

After a Sleep session, the session shows how long it took you to fall asleep: Auralis asks once (with Apple's Health sheet) to read sleep, and finds when the night's sleep began after the session started. If you're new to Auralis, that's asked together with the patterns above, when the review offers them, and if you say not now it isn't asked again on its own. This is read on your iPhone in the same way, kept only on your iPhone (not in iCloud), and never leaves it. When your Apple Watch watched the session for sleep (below), the next morning Auralis compares when it noticed you fall asleep (or that it didn't) with when Health's record says you did, on your iPhone, and uses the difference only to tune when the Watch notices sleep for you. For each of the last 60 such nights, the times involved (when the session began, when the Watch noticed sleep, when Health says it began, when you last spoke in the session) and that setting are kept in a file on your iPhone (not in iCloud or device backups), never sent anywhere, and deleted with Delete All Sessions.

To learn your own heart rate as you fall asleep, after a Sleep session your Apple Watch watched Auralis also asks once (with Apple's Health sheet) to read heart rate, and then, at most every six hours while "Notice when I fall asleep" is on, reads your sleep and heart rate from the last 14 nights in Health (any night your Watch recorded, not only nights with a session): only the heart rate readings from the half hour before each night's first sleep to a quarter hour after it. On your iPhone it keeps just two numbers a night (the median heart rate before and after you fell asleep) and when you fell asleep, in a file on your iPhone (not in iCloud or device backups), used only to tell when the Watch should notice you've fallen asleep, never sent anywhere, and deleted with Delete All Sessions.

If you turn on Wake me for Wake back to bed (Dream incubation aiming for a lucid dream), Auralis reads your sleep stages from the last month, with the same permission, to work out when REM sleep usually comes for you, and sets the alarm for the end of it. The stages are read on your iPhone when the session starts, used only for that night's timing, and neither kept nor sent anywhere. Nothing is recorded during the night. (Until 28 October 2026 the lucid practices could also play a chime in the night, timed the same way; it has been removed.)

Your week. If you allow it there, Auralis reads today's workouts (Apple's Health sheet asks once) to tick a day outdoors: only whether a hike, a ski, a climb, or a walk, run or ride outside lasted 20 minutes or more (Auralis's own Movement workouts don't count). They are read on your iPhone, never kept, and never sent anywhere.

Lucid dreaming. Your answers the morning after a lucid night (whether you were lucid, and, for nights before 28 October 2026, whether you heard the chime), the dream signs you add or hide, and the journal's recurring dream images used as dream signs are kept only on your iPhone (not in iCloud). Your dream signs (since 26 October 2026) are counted on your device from your own words in your dreams, never sent anywhere to be found; the sign you choose as tonight's reality check stays on your iPhone, and a sign you mark Not a sign is kept with your symbol notes, synced with your journal when sync is on, and deleted with the journal. A lucid night's session names up to three of them aloud before you sleep, and tonight's sign after the intention, so those lines are part of the session like the rest of its guidance: kept in its transcript (synced with your sessions when sync is on), read by the model that sums the session up, and spoken by the voice you chose. Reality checks are local notifications scheduled on your iPhone (on or off in Lucid dreaming, or Settings → Reminders). If you ask Auralis to wake you for Wake back to bed, it sets an alarm with Apple's alarm system for that one time.

Apple Watch. If you have the Auralis Watch app, it follows a session started on your iPhone: it runs a mindfulness workout to read your heart rate, shows it, and sends it to the iPhone, where the session can ease the music toward calm when your heart rate climbs. The workout isn't saved to Health, except in Movement (below). About once a minute a reading is kept for the session's review, on that iPhone only, until the session is erased: it isn't part of the session's record, so it never goes to iCloud, even with sync on, and is left out of device backups (the session's log notes only that a reading came, without the number). It's never sent to an AI provider. You can turn the Watch off for sessions, or just the music easing, in Settings → Health & Devices.

The Watch app also has breathing sessions of its own. The first time you start one, the Watch asks (with Apple's Health sheet) to run a mindfulness workout, read heart rate and save Mindful Minutes. During the session the workout keeps it running with your wrist down and shows your heart rate on the Watch; the workout is then discarded, and if you allowed it the session is saved to Health as Mindful Minutes (its start and end time, nothing else). The heart rate readings stay on the Watch. The Watch keeps a count of today's breathing minutes for its own screen and complications. While you breathe, the Watch shows your heart rate and how it has changed since you began (two numbers, such as 74 to 66); it is a heart rate and says nothing about stress. What is kept of a session is described next.

The Watch on its own. The Watch app runs without the iPhone app (an independent Watch app): breathing sessions and sits work, and are saved to Apple Health on the Watch itself, with the iPhone away, off or not even installed. Auralis asks Health for permission on the Watch the first time it is needed. Each session done this way is also kept on the Watch as a small record (its kind, start time, length, the breathing pattern and breath count, and three heart rate numbers: the average and where it began and ended) until your iPhone has it. The Watch then sends it over WatchConnectivity, directly to your iPhone, when the iPhone is next in reach, even if the app isn't open. The iPhone adds it to History as a session "On Apple Watch" (once; a repeat is ignored) and keeps the heart rate numbers on that iPhone only, like any session's: not in the session's synced record, not in iCloud, not in device backups, and never sent to an AI provider. The iPhone doesn't save a second time to Health: only the Watch does. The Watch also sends short lines for the iPhone's Watch log (that a session was saved and queued, with its kind and minutes, never anything said and no heart rate numbers). A record the iPhone never confirms is dropped from the Watch after 30 days. Voice sessions, music, Sleep, dreams, questions and check-ins still need the iPhone, and the Watch says so when it isn't in reach.

Meditation on the Watch. The Watch asks your iPhone which meditation practices your traditions offer (their names and a line each, from your iPhone to your Watch, never off your devices) and tells it which one you chose. A sit on the Watch alone (1 to 60 minutes) keeps time with taps on your wrist and, with Log to Apple Health on, saves Mindful Minutes (its start and end time, nothing else) the same way a breathing session does. During a session on your iPhone the Watch can tap your wrist at its bells; that needs nothing more than the session's status, and bell taps can be turned off on the Watch.

Movement on the Watch. With Log to Apple Health on, a Movement session on your iPhone is recorded by the Watch when it's there: it runs a workout of the practice's kind (walking, yoga, tai chi and so on) and saves it to Health on the Watch with its start and end, the heart rate and active energy the Watch measured, as Apple's Workout app would; the iPhone then saves no workout of its own. If the Watch isn't there or can't confirm in time, the iPhone saves its workout as above. You can also start Movement on the Watch: the first time, Apple's Health sheet asks to save workouts and read heart rate and active energy. The Watch asks your iPhone to begin the guided session; if your iPhone isn't in reach it's a timed workout on the Watch, saved to Health the same way unless Log to Apple Health is off on your iPhone. These readings stay in Health on your devices; Auralis doesn't send them anywhere.

Your iPhone and Watch talk to each other directly over Apple's WatchConnectivity (the same private link Apple uses between paired devices; Auralis has no server in between). The iPhone sends the Watch the running session's name, part and time left (and in Movement the line being said, from the practice's script, and a request to record the workout), and, for the Watch's home screen and complications, today's path day (the path's name and the day's title), the day's quotation from the Library, today's practice minutes and whether Log to Apple Health is on. With App Lock on, a path made from your journal is sent as "Your path", without its day's title, as the iPhone's widgets show it. The Watch sends the iPhone your heart rate during a session, and your requests to start a Meditation, Sleep, Breathwork or Movement session there or to pause, resume or end the running one, and whether it has recorded a Movement workout (or, if it couldn't save one, its kind and times, for the iPhone to save instead). None of it goes anywhere else.

The iPhone also sends the Watch the timing of the next minute or so of breaths and holds (so the Watch can tap them) and your usual bedtime (so the Smart Stack can offer Sleep at the right time). While the Need Help button is on the iPhone's screen, it sends the fixed lines with 988 (when Need Help offers it) or the emergency number, so the Watch can offer them and Ground Me; never your trusted contact. If you ask for Wake back to bed's alarm on your Watch, it sends the alarm's time, and the Watch tells it when the alarm is set (and when you say you're awake). From the Watch you can:

Heart rate and stillness guiding a session (both in Settings → Health & Devices). "Heart rate guides the session" is off unless you turn it on: in Breathwork your heart rate can gently slow the breath. "Notice when I fall asleep" (on by default) uses your heart rate and, during a Sleep session only, how still your wrist is (the Watch's motion sensor; it asks permission first) to let the session end sooner when you seem asleep (nothing more is said, the microphone goes off and the music fades out), and keeps that time on your iPhone (not in iCloud) so the session's review can show it and the next morning can compare it with Health (above). Once Health's recent nights show your heart rate as you fall asleep (above), that is compared with your heart rate during the session too, on your iPhone. Anything you said in the session in the ten minutes before means you were awake: only that you spoke is used, from what the session already heard. The motion readings are reduced on the Watch to a single "how still" number every half minute, sent to your iPhone and not kept. Heart rate and motion stay on your Watch and iPhone and are never sent to an AI provider.

Music

If you connect Apple Music, Auralis uses Apple's MusicKit to play music and read your library or playlists for soundtrack choices. Playback and your Apple Music account are handled by Apple under Apple's privacy policy. Auralis does not share your sessions or journal with Apple Music. If you keep music from another app playing, Auralis reads only what the system reports as playing (title and artist), and can pause or resume it when you ask.

Your music taste. With Apple Music connected, Auralis reads your library playlists on this device (their names, their songs' titles, artists, genres and lengths, whether a song has lyrics, and how often and how lately you played it) and the songs that played in your sessions (which you skipped or loved) into a short summary of your taste: favourite artists, genres, moods, and roughly how energetic and how sung your music is. It's read in the background at most twice a day and kept in one file on your device; it isn't synced or sent anywhere on its own. It's used to suggest your own music for a practice and to choose music in sessions. Settings → Music → Your taste shows all of it, and there you can turn off any playlist or any artist ("Don't learn from this"): its songs, plays and Apple Music signals are then left out of the taste and its summary (so they're never named to an AI model as your taste) until you put them back. That choice is kept on your device with your other music settings.

What Apple Music knows of you. With the library, Auralis also asks Apple Music for your personal recommendations, the songs you've played lately and your heavy rotation (the names, artists and genres of the albums and songs in them), and whether you've loved or disliked your most played songs. They count a little in the same taste summary, less than your own playlists; nothing else is made from them, and they're kept in the same file. If your library is small, Auralis also finds Apple Music's own mood playlists (calm, sleep, ambient, meditation, focus) by a catalog search and plays from them until your own music says more. Nothing new is sent to an AI model: the taste summary is the one described here.

Loving a song. When you tap Remember on a song or say "I love this", Auralis also loves it in Apple Music (through Apple's Music API), as if you'd tapped the heart there. Saying "keep going like this" or "more like this", or a long conversation running out of music, starts an Apple Music station from the song playing; Apple chooses its songs. On headphones with spatial audio, Auralis checks which songs Apple Music offers in Dolby Atmos (from the route your iPhone reports, read on the device) to prefer them in journeys, sleep and Talk with Auralis.

Songs' previews, measured overnight. To know how your songs really sound, Auralis listens to each one's 30-second Apple Music preview on your iPhone: overnight, only while it's charging and on Wi-Fi (never cellular), as a background task the system runs, at most 300 songs a night (songs from your playlists, songs played in your sessions, and songs offered or picked when you asked for music). Each preview is downloaded from Apple, measured on the device (its tempo, energy, brightness, whether it has singing or speech, the instruments heard, and sounds such as rain or the ocean) and deleted. Only those numbers are kept, in one file on your device; nothing about it is sent anywhere except the preview's download from Apple, and the numbers for the songs offered to an AI model when you ask for music (below).

Asking for music. When you ask for music in your own words ("make me a playlist of things you think I'd like", "like my Ott playlist but no vocals"), Auralis searches Apple Music's catalog for the artists and words involved. If an AI model is set up (Auralis Cloud, your own key, or your Mac), it sends that model your taste summary (artist, genre and playlist names, and the numbers above), your request, the session's practice and the song playing, and the titles, artists and genres of the songs to choose from, with their measured tempo, energy, voices and instruments where they've been measured. Never anything else you said, your journal, or your sessions' words. Without a model, or with replies kept Private, the choice is made on your device. A playlist is saved to your Apple Music library only after you've said yes once (Settings → Music).

Each song in a session's review links to Apple Music. For a song that played from your own library or was recognized by ShazamKit, Auralis may look it up once in the Apple Music catalog by its title and artist to link its page; what it finds is kept on your device. Nothing else about the session is sent.

Safety features

During a session, Auralis can offer to call a trusted contact you entered, or emergency services. Nothing is ever dialed or sent without you tapping to place the call, and your trusted contact is stored only on your device (not in iCloud, not with your sessions, and never in a feedback report). What you write under When it's hard stays with your settings, as described above, and is said only by a voice on your device. Auralis is not a medical device and does not provide medical advice.

Children

Auralis is rated 18+ on the App Store and is meant for adults; it is not directed to children, and we don't knowingly collect information from anyone under 18. What we receive at all is limited to what's described above: Auralis Cloud's anonymous daily totals and the counts filed under a hashed subscription id, anonymous performance reports and usage counts (unless you turned them off), a support email you choose to send, and beta feedback a tester chooses to send.

Changes

If this policy changes, we'll update this page and the effective date above.

Contact

Questions: support@helixstudios.app