Privacy Policy

What this is

Told is an AI wellness companion for your inner life. You speak, Told listens and remembers what matters to you, and over time helps you see the shape of what you’ve been holding. This policy explains what we collect, how we use it, and the choices you have.

What we collect

Account information. Your name and email address when you sign in with Google, Kakao, or Apple.

Voice recordings. Audio from your conversations, recorded with your explicit consent at the start of each conversation.

Transcripts and reflections. Text derived from your recordings, including transcriptions, extracted facts, and AI-generated reflections.

Emotional context (inferred from conversation content and rhythm). Told may note emotional context from the meaning of what you share — the content of your words — and from the rhythm of the conversation: how long you pause and how quickly you speak compared to your own usual pace. We do not analyze vocal tone, pitch, or other acoustic characteristics of the sound of your voice to infer emotion. This context is used only for your own features — never for advertising, never shared. See "Emotional context" below.

Usage data. Conversation duration, feature interactions, and technical diagnostics to keep the service running well.

Voice data and biometric information

We record and store the audio of your voice conversations so it can be transcribed and turned into the reflections described above.

No voice model or voiceprint. We do not create a voice model, voiceprint, or any other biometric identifier from your recordings, and we do not use the sound of your voice to identify you.

No sale or commercial use. Your voice recordings and the data derived from them are never sold, leased, traded, or used for any purpose other than providing Told to you.

How your data is processed

Real-time conversation. During a conversation your audio streams through LiveKit (WebRTC infrastructure) to Speechmatics (live speech-to-text). An AI hosted by OpenAI guides the conversation, with deeper reasoning support from Anthropic. Depending on the voice you choose, Cartesia or ElevenLabs converts its replies to speech. If an ElevenLabs voice cannot start, Cartesia supplies a backup voice; if a voice stream fails after starting, OpenAI supplies the final backup. If the AI looks something up on the web, only a short search phrase — never your transcript — is sent to Tavily. OpenAI and Anthropic do not train models on data sent through their APIs, and we have opted out of model training with Voyage AI; we are completing training opt-outs with our remaining voice providers.

Post-conversation processing. After a conversation, the recording is transcribed by ElevenLabs, and your transcript is sent to Anthropic for fact extraction and reflection generation, and to Voyage AI for semantic embeddings that power contextual search. These models do not retain your data after processing.

Emotional context

Told understands emotional context from the meaning of what you share — the content of your words — and from conversational rhythm: pause length and speaking pace measured against your own usual rhythm. Told learns your conversational rhythm over time and uses it to time and shape its responses — for example, giving you room when you pause. This rhythm data is linked to your account and used only to personalize your conversations. We do not analyze vocal tone, pitch, or other acoustic characteristics of the sound of your voice to infer emotion, and no voiceprint is created from your voice. This understanding comes from our own systems and the AI processing described in Third-party AI processors.

Emotional context is used solely for your own features — your Living Portrait and reflections — and to help the AI respond naturally. It is never used for advertising and never shared. It is not displayed to you as emotional labels or interpretations — we do not tell you "you sounded sad."

Notifications and re-engagement

TOLD may send you notifications to surface pattern insights from your Living Portrait — for example, "A pattern you discussed three months ago has resurfaced in recent conversations."

Consent. We will ask for your explicit permission before sending any notifications. You can enable or disable notifications at any time in your account settings.

Preferences. You can control which types of notifications you receive and mark specific conversations as private (excluded from pattern surfacing).

Sensitivity. We apply filters to avoid resurfacing content associated with emotional distress. Recent conversations (less than 30 days old) are never resurfaced.

Third-party AI processors

Your data is processed by the following services, each under a data processing agreement:

• LiveKit — real-time audio transport (infrastructure; carries your voice audio)
• Speechmatics — live speech-to-text during conversations (receives your voice audio in real time)
• ElevenLabs — text-to-speech synthesis for some voice choices (receives the AI's replies), and speech-to-text transcription after a conversation (receives your recorded voice audio)
• OpenAI — conversational AI (receives conversation text) and automatic backup voice (receives the AI's replies)
• Cartesia — text-to-speech synthesis for some voice choices (receives the AI's replies)
• Anthropic — in-conversation reasoning, fact extraction, and reflection generation (receives conversation text)
• Voyage AI — semantic search embeddings (receives conversation-derived text)
• Tavily — web search (receives short search phrases only, never your transcript)
• Datadog — error monitoring, performance diagnostics, and a masked screen recording inside the signed-in app (receives page views, clicks, and your name and email address; never your conversations)
• PostHog — product analytics (receives which screens you open and which features you use, linked to your account; never your conversations)

On this website we collect anonymous usage analytics — which pages are viewed and which buttons are clicked — to understand how people find Told and where the site confuses people. That measures pages, not people: no advertising trackers, and no profile built about you.

Inside the signed-in app, diagnostics are linked to your account so we can fix what breaks for you specifically. That includes a masked screen recording: the text you type and the words on screen are hidden on your device before anything is sent, so what we receive shows layout and interaction rather than content. Your name and email address are attached to these diagnostic records. Your conversations, recordings, and transcripts are never part of analytics or diagnostics, and none of it is ever used for advertising or sold to anyone.

Where your data is stored

Your recordings and all derived data are stored on servers in the United States using Amazon Web Services. Every connection to Told is encrypted in transit, and your voice recordings are encrypted where they are stored.

How long we keep it

Your recordings and conversation data are kept for as long as your account is active. If you delete your account, we will delete your recordings and personal data within 30 days. Anonymized usage statistics may be retained.

Your rights

You can request a copy of your data, ask us to correct inaccuracies, or delete your account and all associated data at any time. California residents have additional rights under the CCPA, including the right to know what personal information is collected and the right to opt out of its sale — we do not sell your personal information.

In-app deletion. You can permanently delete your account and every conversation tied to it from inside the app: open Settings, scroll to Delete account, and confirm. Deletion is immediate and irreversible. Audio storage cleanup begins automatically, and storage copies are fully cleared within 30 days.

Biometric data rights. If you are a resident of Illinois, Texas, Washington, or another state with biometric privacy laws, you have additional rights regarding your voice data, including the right to be informed before collection, the right to consent to or refuse collection, and the right to have your biometric data destroyed. Contact us at hello@told.one to exercise these rights.

Children

TOLD is designed for adults. Only adults (18+) may create accounts and use the service. TOLD is not suitable for minors.

Disclaimers

Wellness, not psychotherapy. Told is a wellness companion. It is not psychotherapy and is not a substitute for licensed mental health care. If you are in crisis, please contact a qualified professional or your local emergency services.

Recording. Voice conversations are recorded so Told can listen, remember, and be ready for next time. You can delete any conversation at any time from Settings: it disappears from your view right away, and its voice recording is removed when you delete your account. See "How long we keep it" above for the full retention policy.

Text mode. When you choose text instead of voice, you are talking to the Told text companion. This is a different experience from the voice conversations described above and uses a different processing pipeline.

How the AI is guided

Told's AI follows a written set of instructions every time you speak with it. We share the gist here so you know what to expect.

What the AI is asked to do. Listen, reflect what it heard, and ask open questions. Remember what matters across conversations. Keep the conversation focused on what you brought up — not on advice or interpretation.

What the AI is asked NOT to do. No diagnosis. No treatment plans. No clinical labels. No "you sound sad" emotion call-outs. No personality reads. If you bring up a crisis — thoughts of harming yourself or someone else — the AI will pause, name its limits, and point you to a real person (such as the 988 Suicide & Crisis Lifeline in the United States).

Memory and pattern observations. Told remembers what you tell it. Over time it may notice patterns across conversations and mention them — at most once per conversation, stated as observation, not interpretation. You can delete any conversation at any time from Settings; it disappears from your view right away, and its voice recording is removed when you delete your account.

Quiet by design. The AI carries a lot of context about you but speaks sparingly. Long silences are allowed. The goal is for you to feel heard, not analyzed.

Changes to this policy

We may update this policy from time to time. If we make meaningful changes, we will notify you by email or through the app before they take effect.

Contact

Questions or requests? Reach us at hello@told.one.