Privacy Policy

This policy explains what data Agentele collects, why, and what control you have over it. The short version: your conversations are yours, we store them to make your agents work, we never sell your data, anything you ask one agent to keep private stays with that agent, and you can export or permanently delete everything from Settings.

1. What we collect

• Account data — email address, display name, timezone, and authentication records (including Google sign-in identifiers if you use Google OAuth).
• Agent data — the agents you create: their personality, schedules, portraits, and the memory facts they distill from your conversations.
• Conversations — messages, voice-call transcripts, and photos you attach, stored so agents keep context and memory. Photos are stored as files with long, unguessable web addresses; anyone given such a link can open it, so treat an attached photo as shareable rather than private.
• Usage and billing — metered usage events (tokens, minutes, cost), your plan and its expiry, credit balance, and gift-code redemptions.
• Technical data — standard server logs and push-notification subscriptions for the devices you enable.

2. How your data is used

• To run the product: generating agent replies, voice, portraits, memories, and scheduled proactive messages.
• To apply your plan's fair-use limits and meter usage.
• To secure the Service and prevent abuse (rate limits, approval queue, and the automatic safety check described in §6b).
• To email you about major product updates, a few times a year at most — every one has a one-click unsubscribe link, and you can switch them off any time in Settings → How they reach you.
• We do not sell your data, show you ads, or use your conversations to train our own models.

3. Who processes it

• Supabase — database and authentication, hosted in the EU (Frankfurt, eu-central-1). All data is protected by row-level security: only your account can read your rows.
• Vercel — application hosting and AI Gateway routing.
• AI model providers (Anthropic, OpenAI, and others via the gateway) — receive conversation content to generate replies, transcriptions, speech, and images. They process it as vendors under their API terms; API traffic is not used to train their consumer models.
• Resend — sends our emails: your agents' messages when you have no notifications on, invitations to shared agents, and the occasional product-update email.
• Push services (Google Firebase Cloud Messaging for the Android app, and your browser's own push service) — deliver notifications and incoming calls to your device.
• Payment processors — when card checkout is enabled, payments are handled by a merchant of record (e.g. Paddle or Lemon Squeezy); we never see or store your card number.

4. Bring-your-own-key (BYOK)

If you add your own provider API key, it is encrypted at rest with per-record envelope encryption and is only decrypted in memory to call your provider. It is never written to logs and never shown back to you in full.

5. Retention and deletion

• Your data is retained while your account is active.
• Safety records (§6b) are kept for 180 days and then deleted automatically — sooner if you delete your account.
• Export — Settings → download a full JSON export of your agents, memories, and conversations at any time.
• Delete — Settings → delete account permanently removes your profile, agents, conversations, memories, schedules, usage records and uploaded files. This is immediate and irreversible.

6. Things you ask an agent to keep private

Anything you tell one agent to keep between the two of you stays with that agent. So does anything that reads as obviously sensitive — health, medication, drug or alcohol use, sexuality, mental health, money trouble, legal matters, relationship trouble — which is marked without you having to ask for it.

• It does not reach your other agents. A marked fact is left out of briefings sent between agents, group calls, the mind map, task extraction, search across your agents, proactive messages and push notifications, and anything an outside app you connected can read. This is enforced in code at each of those exits, not by asking the agent to be careful.
• It asks before including one. If you ask an agent for something that would contain a marked fact — a summary, a handover, a note for someone else — it writes the rest and then asks whether to include it, giving only the number of things being held back, never what they are.
• You decide what is marked. Every fact an agent holds, and every mark on it, is listed on that agent's memory page with the reason it was marked. You can add a mark, remove one, edit the fact or delete it.
• What it is not. It is a rule about what your agents pass to each other and to other apps — not a second copy of the data held somewhere else, and not encryption we could not read. Marked facts are stored like any other data described in this policy. A per-agent memory download leaves them out unless you ask for them; the full account export in Settings includes them, clearly labelled, because it is your own complete data.

6b. Automatic safety checks

To keep Agentele within the law and our terms, every message you send to an agent or to Aria — typed, or transcribed from a voice call — and every portrait description is checked automatically as it is sent. A small AI model (run by the same AI providers listed in §3) and a few simple rules look for the uses listed in the terms, §5. No person reads your messages to do this.

• When nothing is wrong — the overwhelming majority — nothing is kept about the check and nobody is told anything.
• When a request clearly crosses a line, the agent declines it, and we keep a short safety record: your account, the agent and conversation it happened in, the category (for example "illegal drugs"), how serious it was, the check's one-line reason, and an excerpt of your message of at most 300 characters. The operator of Agentele may review that record, and the conversation around it, to decide whether to act — for instance to dismiss it as a mistake or to suspend the account. Borderline questions are answered, not recorded.
• How long — safety records are deleted automatically after 180 days, and immediately if you delete your account.
• Self-harm is different. If you tell an agent you are thinking about hurting yourself, it answers with care and offers a crisis line. Those conversations are never reported, recorded as a safety event, or shown to anyone. The only trace is an anonymous daily count of how often support was offered — a date and a number, with no link to any account, agent or words.

7. We do not study you

We do not compute statistics across accounts, and we do not read or analyse what you or your agents say in order to improve the product. (The automatic safety check in §6b exists to keep the Service lawful, not to study you; its only cross-account number is the anonymous support count described there.) An earlier version of this policy described a statistic about agent settings, kept only above 25 accounts, with an opt-out; that feature was removed in September 2026, it never stored a row, and its table has been deleted. There is nothing to opt out of.

8. Cookies

We use only the cookies required for sign-in sessions. No advertising or cross-site tracking cookies.

9. Age

The Service is for adults 18+. We do not knowingly collect data from minors.

10. Changes and contact

We will announce material changes in the app. Privacy questions or requests: support@agentele.ai.