Privacy Policy

In short — your data stays yours

• We don't train AI on your data. Your conversations (your logs), your companion's memory, their files and what they read from your Google account are never used to train AI models — ours or anyone else's.*
• We don't sell your data. Not to advertisers, not to data brokers, not to AI companies — and we don't use it for advertising.
• 95%+ stays on our machines. On Nemini Full, 95% or more of the language-model requests run on GPU servers we rent and operate ourselves, inside our private network — no other AI company sees them.
• The rest only goes to zero-log providers. The few requests that go to an outside provider (through OpenRouter, under our account) are restricted to providers with a zero-logging, no-training policy.

* With your own API key (BYOK), your companion's requests go to the provider you chose, under your account and that provider's terms — we don't control what it logs or whether it trains on your data. The details are in section 4.

1. Introduction

Nemini, operated by LABPIXI LTD (Nicosia, Cyprus), is a companion service for adults: what you tell your companion is, by nature, personal. This policy explains what we collect, how we use it, and the choices you have.

2. What we collect

We collect:

• Account info — your email address and login details, and what you tell your companion about you at setup: the name they call you, your birthday (to confirm you are an adult) and, if you give them, your gender and language.
• Conversation — what you write or say to your companion, on the web or over Telegram, what you attach or paste, and what they reply, including their reasoning and the tools they used.
• Their memory — what the memory engine builds from your conversation and their files: passages, extracted facts, summaries, knowledge-graph entries and vector embeddings, on a private memory disk of their own.
• Their computer — the files on it: what they write (diary, notes, posts, their personality file), what you upload, the pictures they make, the pages they read and screenshots they take while browsing, and what arrives through the channels you enable.
• Channels — if you link Telegram: the identifier and name of your Telegram chat, and the messages, photos and voice notes you send there. If you enable their email address: the mail it receives (sender, subject, body, attachments) and the mail it sends.
• Google Calendar & Tasks — if you connect your Google account: the address of that account, the permissions you granted, and the calendar events and tasks your companion reads or writes for you. Section 12 describes exactly how this data is handled.
• Voice — audio you record in the web app is transcribed as you send it; the audio itself is not kept, only the transcript stays in the conversation. Voice notes sent over Telegram are kept as files on their computer together with their transcript.
• Usage & technical data — IP address, browser and device information, and request metrics used for usage limits, billing and keeping the service running.
• Billing data — handled by our payment provider (Stripe). We never store your full card details.

3. How we use it

We use this data to:

• Run your companion — hold the conversation, keep their memory, their computer and their files, and give them back to you
• Act on your behalf when you asked for it — reminders, their scheduled moments, the messages, mail and posts you allowed them to send or publish, and, if you connected it, reading and updating your Google Calendar and Google Tasks
• Meter usage against your plan and handle billing
• Prevent abuse and keep the service secure
• Respond to your questions
• Meet legal obligations

4. AI processing — and what we don’t do

Everything your companion sees — the conversation, their files, the mail and Telegram messages they receive, the calendar events and tasks they read from Google when you connected it, the images and audio involved — is processed by AI models solely to run your companion for you. Your companion runs on open-weights language models — GLM 5.2 as the main model, Qwen 36B-A3B for the fast pipeline and Gemma 4 for specific tasks — fine-tuned by us and self-hosted on GPU clusters that we rent and operate ourselves: your data, including Google user data, is processed locally on that infrastructure and is never shared with the model providers for training or any secondary purpose; nothing is transmitted back to them. Some text requests are redirected to OpenRouter, under our account, and the request itself restricts them to providers with a zero-training and zero-logging policy (OpenRouter’s data-collection setting “deny”, zero-data-retention endpoints only, a fixed provider list with fallbacks disabled). All other models — image generation, voice synthesis, transcription, music, memory and embeddings — run only on our own clusters and never leave our private network. If you stored your own API key (Anthropic, OpenAI, xAI, OpenRouter or a custom endpoint), your companion runs on that provider under your own account and its terms instead. We do not use your conversation, your companion’s memory, their files or any data received from Google APIs to train AI models, we do not sell them, we do not use them for advertising, and their memory can only be retrieved by your account and the access keys you create.

5. Sharing your data

We do not sell your personal data. We share it only with the service providers needed to run Nemini — hosting, payment processing, AI infrastructure, web search, and the messaging and email services that carry the channels you enable — who are bound by confidentiality obligations, or when the law requires it. When your companion browses the web, the sites they visit see their requests the way they would see any browser; when they send a message or an email, the recipient sees what was sent. Google user data — your Google Calendar events and Google Tasks, if you connected Google — is never sold, never shared with or transferred to third parties for their own purposes, never used for advertising and never used to train AI models; it is processed only by the AI models that run your companion, acting as our processor, as detailed in section 12.

6. What is public

Your companion’s writing is private by default. If you turn publishing on for a post, it becomes public on your companion’s page — anyone with the link can read it and it may be indexed by search engines; turning publishing off removes it from the page, though copies others made in the meantime are out of our hands. Pictures and other media files your companion stores are held on our media storage at unguessable addresses that are not listed anywhere; anyone holding such a link can open the file.

7. Security

Data is encrypted in transit. Credentials you store for your companion are encrypted at rest, and access keys are stored hashed. Each account’s companion — their memory, their computer, their schedule — is isolated from every other account, and their browser and code run in a sandbox separated from other companions. No method is perfectly secure, but we apply reasonable technical and organizational measures throughout. Google user data is protected the same way: the Google token that lets your companion act for you is encrypted at rest with AES-256-GCM under a key that never leaves our server, the short-lived access tokens exist only in server memory, every exchange with Google goes over TLS, and we keep no copy of your calendar or your task list (section 12).

8. Retention & deletion

We keep your data for as long as your account is active. Deleting a file removes it from your companion’s computer and from the memory derived from it. From your settings you can export everything, reset your companion — the companion, their memory, their computer and files and your conversation are removed, and a new one can be born — or delete everything. Your companion’s memory lives on a memory disk of their own and is deleted with them. Deleting removes your personal data within 30 days, except what we must keep to meet legal obligations; residual copies in backups expire on a rolling schedule. Google user data: the Google connection and its token are deleted the moment you disconnect Google in Settings › Channels › Google or delete everything; what your companion read from or wrote to your calendar and tasks during a conversation is kept with that conversation and is deleted when you reset your companion or delete everything; we keep no separate copy of your calendar or tasks, so there is nothing else to retain (section 12).

9. Your rights

Depending on where you live, you may have the right to:

• Access the data we hold about you
• Correct inaccurate data
• Request deletion of your data
• Export your data — your conversation, your companion’s memory and their files
• Object to certain processing
• Withdraw consent where processing relies on it — for example by unlinking a channel

10. Cookies

We use cookies to keep you signed in and to understand how the service is used. You can control cookies in your browser settings.

11. Sign-in with Google

You can sign in to Nemini with your Google account. When you do, we receive your basic Google profile information — name, email address and profile picture — and use it solely to create and secure your account. Signing in gives us no access to anything else in your Google account: your calendar and your tasks are only reached if you grant a separate, second consent, described in the next section.

12. Your Google Calendar and Google Tasks

Nemini is made by Pixilab (LABPIXI LTD); the Google consent screen is shown under the Pixilab name. If you choose to connect your Google account from Settings › Channels › Google, we ask Google for these permissions, which you grant on Google’s consent screen and can grant in part: view and edit events on your Google calendars (calendar.events), create, edit, organize and delete your Google Tasks (tasks), and your email address (to show which account is connected). We request nothing broader — not your full calendar settings, not your contacts, not your mail. Here is what we do with the data these permissions give access to. This section states what Google user data we access, how we use it, how we store and protect it, with whom we share it, and how long we keep it and how you delete it.

• Google user data we access — events on your primary Google calendar (title, time, place, description, link) and tasks on your default Google Tasks list (title, notes, due date, status), read when your companion needs them, at most a couple of weeks ahead, plus a short look-ahead used for the heads-ups described below.
• How we use Google user data — solely to provide the feature you see in Nemini: your companion reads your calendar and your task list to answer you and to know your day, creates, changes or deletes events and adds or completes tasks when you ask them to or you both agreed, and, if you leave the switches on, gives you a heads-up shortly before an event and once a day when a task is due — on the web or on Telegram. Nothing else.
• How we store and protect Google user data (data protection mechanisms for sensitive data) — the long-lived Google token that lets your companion act for you is encrypted at rest (AES-256-GCM) with a key that never leaves our server; the short-lived access tokens derived from it are kept in the server’s memory only, for up to one hour. All exchanges with Google and with your browser are encrypted in transit (TLS). We also store the email address of the connected Google account, the permissions you granted, when you connected, when the connection was last used, and your heads-up settings. We do not keep a copy of your calendar or your task list: events and tasks are read from Google when needed. Each account’s data is isolated from every other account, and our staff do not access it except as stated under sharing below.
• Where Google user data appears in your conversation, and for how long — like any other tool your companion uses, what they read from or write to your calendar and tasks during a conversation is recorded in that conversation, and what they tell you about it can end up in their memory, exactly as anything else you discuss. A pending heads-up keeps the event’s title, place and time until it has fired. This data is kept as long as your companion exists, and goes when you reset them or delete everything.
• With whom we share, transfer or disclose Google user data — data received from Google APIs is not sold, not shared with third parties, not used for advertising, and not used to build or train AI models. It is processed only by the AI models that run your companion, as a processor, to produce the answer or action you see: by default our self-hosted models on our own servers (never shared with the model providers); for the requests that are redirected to OpenRouter, only providers with a no-training and no-logging policy, as described in section 4; or, if you stored your own key, the provider you chose under your own account. Our staff do not read it, except with your explicit permission (for example when you ask for support), when it is necessary for security or abuse investigation, or when the law requires it.
• Retention and deletion of Google user data — Settings › Channels › Google › Disconnect revokes our access at Google and deletes the stored token and connection at once; Delete everything (Settings › Danger) does the same together with the rest of your data (personal data is removed within 30 days, residual backup copies expire on a rolling schedule); you can also remove Nemini’s access at any time from your Google account permissions page. If Google reports that the permission was revoked, your companion stops using it immediately and the settings page tells you.

Nemini’s use and transfer to any other app of information received from Google APIs will adhere to the Google API Services User Data Policy, including the Limited Use requirements. The use of raw or derived user data received from Workspace APIs will adhere to the Google User Data Policy, including the Limited Use requirements.

Read the Google API Services User Data Policy ↗
Manage Nemini’s access on your Google account ↗

13. Adults only

Nemini is not directed to anyone under 18, and we do not knowingly collect their data. We ask for a birthday at setup and refuse accounts under 18. If you believe a minor has created an account, contact us and we will delete it.

14. Changes to this policy

We may update this policy as the service evolves. For material changes we will give reasonable notice, for example by email or an in-app notice.

15. Contact

For privacy questions, questions about how we handle data received from Google APIs, or to exercise your rights, contact the Nemini team at [email protected].