Privacy, plainly.

The app opens to people from the waitlist, in small groups. This policy describes it as you will use it.

Who we are

Soundthought is operated by Innerworks UG (haftungsbeschränkt), a small company based in Germany, built and run by Soundthought's founder (“we,” “us”). The company is registered at the Amtsgericht Frankfurt am Main under HRB 144843; the full legal details are in the Impressum. This policy covers the Soundthought mobile app and this website. For anything in it: hello@soundthought.ai.

This website

Visiting

The site is served by Cloudflare. Like any web server, it momentarily sees standard technical data (your IP address and browser type) to deliver pages and fend off abuse; we don't use it to identify or profile you. The fonts are hosted on our own domain, so opening this site doesn't hand your visit to Google or any other font service. Nothing on these pages loads from anywhere else. Your light/dark theme choice is saved in your own browser, on your device. It never reaches us. The site sets no cookies.

Counting visits

Cloudflare, which serves the site, gives us aggregate traffic numbers, like how many visits a day it answers. No script runs in your browser to count you. We see counts, never people.

The waitlist

If you join the waitlist, we store your email address in our database (hosted in the EU). It is used for exactly three things: one email asking you to confirm it, one welcome note once you do, and your invitation when it's your turn. We also note when you confirmed, so we can show you asked to join. If you don't confirm within 7 days, we delete the address. Nothing else, and it is never sold, shared, or used to profile you. Every email we send carries a link that removes you, and clicking it deletes the entry outright. You can also write to hello@soundthought.ai. You do not have to explain why, and there is no button we have hidden from you.

The app: what we collect, and why

Account basics. Your email address, sign-in credentials, and the name you'd like to be called.

What you share in conversation. When you speak or type to Soundthought, your words are processed to generate a response. What persists afterward is the thread of the conversation: transcripts, summaries, the “memories” the app keeps for continuity, how far you have got in the sessions, and brief tone notes. Those last ones are worth stating plainly: Soundthought listens to how you sound as well as what you say, and it keeps short signals like “sounded tired” attached to the thread, so a later conversation can pick up where the last one left off. All of it lives in our EU database, tied only to your account and visible to you in the app. The Security section below is honest about what “encrypted” does and doesn't mean here.

Your voice: processed, never kept. Spoken audio is streamed, transcribed, and read for tone in real time, then discarded. This is an engineering rule in Soundthought's architecture, not a preference: audio is never written to storage, and your voice is never used to identify you.

Journal entries and check-ins. Shown only to you in the app, and read by Soundthought's systems for one purpose: giving your conversations continuity. Voice journal entries follow the same rule as conversation: the transcript and tone tags are kept, the audio is not.

Basic diagnostics. Crash reports and performance signals, so we can keep the app working. These are engineered to exclude your content (no transcripts, no names, no journal text, no emotion scores) and are deleted after 90 days.

How the AI works with your words

Every model Soundthought uses runs on infrastructure we operate. The conversation model, voice transcription, tone reading, and Soundthought's spoken voice are all self-hosted: your words are never sent to a third-party AI service, and there is no outside AI company in the loop. No third party ever trains on what you share. Soundthought's own models can learn from conversations only if you explicitly opt in, and only de-identified. The section below explains exactly how.

The legal basis for each use

• Joining the waitlist: your consent (GDPR Art. 6(1)(a)), which you give by confirming your address. The link in any email withdraws it.
• Your account and the app: our agreement with you (Art. 6(1)(b)).
• What you say and write in the app, including tone notes: your explicit consent (Art. 9(2)(a)), because what you share can touch on your health, which the GDPR treats as a special category of data.
• Helping Soundthought learn: a separate explicit consent, off by default.
• Keeping the site and app secure and working: our legitimate interest (Art. 6(1)(f)), which you can object to.

Withdrawing consent never affects what happened before.

Who touches your data

No analytics brokers, no ad networks, no data buyers. The complete list of processors is:

• Supabasedatabase and sign-in, hosted in the EU
• Amazon Web Servicesholds the master key that locks the key unique to you; it never sees your content
• Modalthe computers Soundthought's self-hosted AI runs on; processing is transient
• Upstashbrief session coordination; holds no conversation content
• Sentrycrash reports, built to contain no personal content
• Cloudflareserves this website and routes our email
• mailbox.orgour email inbox, where mail to hello@ is kept; servers in Germany
• Resendsends the waitlist emails; EU servers, tracking off

Each processes data only on our instructions, under a data-processing agreement.

Where your data lives

Your stored data lives in the European Union. Soundthought is available worldwide, and the computers that run Soundthought's AI are provided by Modal, a US infrastructure company. Processing there is transient (nothing is stored on them), under EU-approved safeguards: a data-processing agreement with standard contractual clauses. Wherever you are, the same promises in this policy apply to you.

Several of our providers are companies based outside the EU, in the US and Singapore. Where one of them can access data from outside the EU, the transfer is covered by the EU-US Data Privacy Framework or by standard contractual clauses.

What we will never do

• Sell your data, or share it with advertisers or data brokers.
• Show you ads.
• Use your conversations for marketing.
• Train AI on your words without your explicit opt-in, or with anything attached that could identify you.
• Write your words into system logs.
• Contact anyone on your behalf. Crisis lines are always one tap away in the app, and it puts them in front of you if a conversation turns that way. It does not call or notify anyone.

Helping Soundthought learn, only if you say yes

Soundthought's own models can get better at listening, but only from freely given help. If, and only if, you explicitly opt in, excerpts of conversations may be used to improve Soundthought's own AI. Before anything is used, it is de-identified: kept apart from your account, and stripped of your name, your email and details that could identify you. This is off by default, entirely optional, and never a condition for using Soundthought. Change your mind any time: opting out stops future use and removes what you've contributed within 7 days. What's contributed trains Soundthought's models and no one else's. It is never shared, sold, or passed along.

Your controls

See what it knows. The app's “What I remember” screen shows every memory Soundthought holds about you.

Forget anything. Remove any individual memory, journal entry, or saved moment, or tell Soundthought to forget everything it has learned about you. Removal is an immediate hard delete from our systems, not a hidden archive. Deleting a journal entry also deletes any AI memory that was derived from it. One thing that forget-everything deliberately leaves alone: the record of your own work, meaning your sessions, the exercises you practised, and how far through you got. That is your history rather than something Soundthought concluded about you, and deleting your account removes it along with everything else.

Delete your account. Available directly in the app. This immediately and permanently erases your account and all associated content: there is no retention window, no “deactivated” state, and no way for us to undo it.

Export. You can request a copy of your data at hello@soundthought.ai.

See who opened it. Every time your content is decrypted, that is recorded. You can ask us for that log and read it yourself. Most of this policy asks you to take our word for something. This is the part you can check.

If you are in the EU/EEA or UK, these controls correspond to your rights of access, rectification, erasure, portability, and objection under the GDPR, and you may lodge a complaint with a supervisory authority; in Germany, that is the data-protection authority of your federal state. If you are anywhere else: we don't gatekeep by geography. The same controls are yours too.

Retention

Your content stays only as long as your account exists, or until you delete it, whichever comes first. Diagnostic data is deleted after 90 days. One honest footnote: like any responsible service, we keep encrypted database backups for disaster recovery. Your content is encrypted with a key unique to you, so deleting your account destroys that key and makes every remaining copy unreadable, backups included. There is no waiting for a backup cycle to turn over, and backups are never used to restore anything you have deleted.

Security, including the part most policies won't say

Everything travels between you and Soundthought over encrypted connections (TLS). What you share, and everything Soundthought keeps from it, is stored encrypted with a key unique to you, never in plain text, and the disks underneath are encrypted too. When the AI needs your words, for a reply, a summary or a memory, our servers decrypt them in memory, only for that moment, and keep nothing unlocked afterwards.

Here is the honest part: Soundthought is not end-to-end encrypted. The AI has to read your words to respond to them, so our systems can decrypt them, which means, technically, we could read them too. We treat that access as something to minimize, not use: access to production systems is restricted and logged, the AI's memory reads are recorded in an audit trail, our system logs never contain your words, and voice recordings don't exist at all. No system is perfectly secure, but Soundthought is built so that even in the worst case there is as little as possible to find. The safest data is the data that was never kept.

Who Soundthought is for

Soundthought is for adults. It is not directed at anyone under 18, and we don't knowingly collect data from anyone under 18. If we learn we hold a child's data, we will delete it.

A note on what Soundthought is

Soundthought is a journal for reflection, not a medical service. It does not create medical records, and this policy is not a substitute for the protections that apply to healthcare providers. If you are in crisis, please reach for humans: in the US, call or text 988, or text HOME to 741741. In Germany, TelefonSeelsorge listens at 0800 111 0 111, free and anonymous. Anywhere else, findahelpline.com lists lines by country, and your local emergency number is always right.

If Soundthought ever has to stop

Soundthought is one person's company, so it is fair to ask what happens to your words if it does not make it. Here is the commitment. If Soundthought has to close, you will be told at least 30 days before sessions stop, and your data will stay exportable for at least 90 days after that. When that window closes, everything is erased from our servers. Your data will not be sold, and it will not be handed to whoever buys the wreckage. It is not part of the furniture.

The honest limit on that promise: it assumes we are in a position to give notice. No small company can rule out the sudden version, and we would rather say so than pretend the plan covers everything.

Changes to this policy

Soundthought is a small company with one person behind it, and this policy may change as the product grows. If anything changes in a way that matters, we will tell you plainly, in the app and by email, before it takes effect. We won't bury it.