AtmanAI is a private, offline AI assistant. Chat runs on your phone. There is no account, no cloud language model, and no analytics SDK on the chat path. This page is the public policy Google Play and App Store listings need, and the copy Settings will open.
This policy describes how the AtmanAI mobile application (“AtmanAI”, “the app”, “we”) handles information. AtmanAI is published by an independent developer. We do not operate a user account system, a chat backend, or a cloud inference API for this app.
For questions about this policy, use the support email published on the AtmanAI store listing, or write to atmanaiapp@gmail.com.
Prompts, replies, and chat history never leave the phone for inference. Models run locally with ExecuTorch after you download them.
The network is used to download model files and to look up file sizes on Hugging Face — not to send conversations.
No login. No API key. No ads. No crash or analytics vendor in the current release. We do not sell data.
In the current release we do not:
Play Console “Data safety” treats “collect” as data that the app or the developer’s servers receive. Local files on your phone are not collection by us. Hugging Face receiving a download request is described in section 6.
The app stores the following in app-private storage (and, on iOS widgets, an App Group on the same device):
| Data | Why it is there |
|---|---|
| Conversation threads | Prompts, replies, titles, timestamps so you can continue a chat. Stored via on-device persistence (AsyncStorage / app files). Not uploaded. |
| Model files | Quantized .pte (and related tokenizer / runtime files) you choose to download. Needed for offline inference. |
| Download state | Progress, pause/resume, destination path, so a large download can survive leaving the screen. |
| Settings | Onboarding-seen flag; whether in-app download notifications are enabled. |
| Catalog size cache | Hugging Face file-size results cached locally (about 24 hours) so the catalog does not re-fetch on every open. |
| Installed-model index | Which models are on disk and which one is active. |
| Notification helpers | Whether permission was already requested; short local records used to keep download notifications accurate. |
| On-device system prompts | Short task presets (chat / code / explain) that steer the local model. They are shipped in the app, not fetched from us at chat time. |
| Clipboard (optional) | Only if you copy a message. Content goes to the system clipboard you already use; we do not sync it. |
Inference (token generation) happens on-device. Streaming replies, “think” traces, markdown, and code blocks are rendered locally. Device-fit checks (RAM, free storage, large-model warnings) use information the OS already exposes to the app; that information is not sent to us.
The app declares Internet access because some features need it. Chat itself does not.
| Use | What is sent / received |
|---|---|
| First-time model download | HTTPS GET of the model file(s) you selected, typically from Hugging Face. Your device IP, user-agent, and the file URL are visible to that host the way any download is. Chat text is not attached. |
| Catalog file sizes | HTTPS request to Hugging Face’s public model API (huggingface.co/api/models/…) to show real byte sizes. Cached on device. |
| Connectivity check | The OS network library (NetInfo) is used to know whether a download can start and to show an offline state. That is a local reachability check, not a call to our API. |
| Background download | Android may keep a foreground/data-sync service, wake lock, and progress notification so a multi-hundred-MB file can finish after you leave the screen. The transfer is still the model file, not your chats. |
After a model is on the device, generation works offline. Opening the app, reading history, switching models already downloaded, widgets, and Settings do not require a connection.
Your carrier or Wi-Fi provider can see that you downloaded a large file. That is ordinary network metadata, not something we collect.
AtmanAI is not affiliated with Hugging Face, Meta, Qwen, or other model authors. The catalog points at public model repositories. When you download or when the app asks for file sizes, Hugging Face (or whichever host serves the file) is a third party. They may process IP address, timestamps, and which artifact was requested under Hugging Face’s privacy policy (or the equivalent policy of another host, if a URL is not on huggingface.co).
The app does not log you into Hugging Face. It does not send your AtmanAI chats to Hugging Face. Model licenses, acceptable-use rules, and weights are those of the model publisher — see the Terms.
Release builds request only what the product needs:
| Permission | Reason |
|---|---|
| Internet | Model download and Hugging Face size lookups. |
| Notifications | Optional alerts when a background download finishes or to show progress. You can refuse or turn this off in Settings. |
| Vibrate | Haptic / notification feedback. |
| Network state | Know whether a download can run. |
| Foreground service / data sync, wake lock, user-initiated jobs (Android) | Keep a model download alive when the app is not in the foreground. |
| Boot completed (Android) | Notification library restore after reboot — not used to phone home. |
Overlay (draw over other apps) and legacy read/write external storage are blocked on release. Debug builds may still use overlay for the React Native error screen; that is not in the Play production AAB.
Onboarding may ask for notification permission so you can see download progress in the system tray (pause / resume / cancel). This is optional. Settings can turn AtmanAI notifications off; you can also revoke permission in system settings.
Notification copy is about download status (model name, progress). It is not a channel we use to message you from a server. There is no push-notification backend for marketing.
Optional widgets (small / medium / large on Android and iOS) show glanceable status: app name, whether a model is installed, a count of conversations from today, and a “new chat” control. They do not render your prompt or reply text.
On iOS, that snapshot is shared with the widget extension through the on-device App Group
group.com.akashdeepvt.localai. On Android, the widget process reads the same local snapshot.
Widgets are not a network feature. Taps open the app via the atmanai:// deep link on the device.
On Android, AtmanAI opts out of Auto Backup (allowBackup="false").
Google’s device backup will not copy this app’s chats, settings, or model files into
your Google account. Uninstalling removes app-private data on that phone. We still do not receive any backup —
there is none of this app in Google backup.
iOS is different: if you turn on iCloud or computer backups in Apple’s settings, iOS may include app data in your backup. That is an Apple product on your account, not an AtmanAI server. Turn iCloud backup off, or exclude this device, if you want chats only on the handset.
AtmanAI is a general-purpose generative-AI chat app. It is not directed at children under 13 (or the equivalent minimum age in your country). We do not knowingly collect personal information from children. Because we do not operate accounts or a chat backend, we do not have a children’s profile database. Parents who believe a child has used the app on a shared device can uninstall it or clear app data.
Local models can generate text that is unsuitable for minors. That is a property of the model weights you download, not of a moderated online service.
We do not run a cloud service that stores your chats, so we are not processing those conversations as a hosted SaaS provider. On-device processing is performed on hardware you control, at your direction.
Where a download request to Hugging Face involves personal data (typically an IP address), that processing is necessary to provide the feature you requested (installing a model) and is performed by that third party as an independent controller of their logs.
We do not sell or share personal information for cross-context advertising. We do not use sensitive categories of data. If a future build adds opt-in diagnostics, this policy and the store Data safety form will be updated in the same release before that binary ships.
Catalog tabs for Vision, Speech, Create, and Tools are locked. They describe possible on-device image, speech, generation, and tool-calling features. They do not currently enable the camera, microphone, or photo library. If a later version uses those sensors or sends new kinds of data off-device, we will update this policy and store disclosures first.
We may update this page when the app’s behavior changes (for example if opt-in crash reporting is added). The “Last updated” date at the top will change. Continued use after an update means you are using the app under the revised policy. Material changes that involve new collection off-device will also be reflected in Google Play Data safety / Apple privacy nutrition labels for that version.
Privacy questions: atmanaiapp@gmail.com (same address as the Play / App Store support email).
We cannot retrieve or delete a chat from our side — we never received it. Use Settings → Clear all chats, delete models, or uninstall on the device.
These terms govern your use of AtmanAI on Android and iOS. By installing or using the app, you agree to them and to the Privacy policy above. If you do not agree, uninstall the app.
AtmanAI lets you download compatible on-device models and chat with them locally. There is no subscription in the current version, no in-app account, and no cloud chat service. Store listing copy that says the product is private and offline refers to inference and history — the first model download still needs a network connection and enough free storage.
Features marked Coming soon are not included. Expo Go and web builds are not supported. Physical devices are required for real inference performance.
We grant you a personal, non-exclusive, non-transferable, revocable license to install and use AtmanAI on devices you own or control, for your own use, in accordance with these terms and applicable law. You may not reverse engineer the app except where the law allows, redistribute the binary, or use our name or mark in a way that suggests endorsement.
Model weights, tokenizers, and names (for example Qwen, Llama, SmolLM, LFM, and others listed in the catalog) belong to their authors and are subject to those authors’ licenses and acceptable-use policies on Hugging Face (or the hosting site). ExecuTorch / PyTorch runtime components are subject to their own licenses. Downloading a model is your choice; you must comply with that model’s license, including any restrictions on commercial use or prohibited categories of use.
We do not warrant that a given model will fit your phone, finish downloading, or run without out-of-memory errors. Large-model warnings in the UI are guidance, not a guarantee.
You agree not to use AtmanAI to:
Because inference is local, we cannot remotely disable a particular chat. If you violate these terms, your license to use the app ends and you must uninstall it.
You are responsible for device storage, RAM, battery, and network charges for large downloads. Incomplete downloads are not treated as ready models. You can delete models and clear chats from the app. Android Auto Backup does not include this app. iOS backups follow Apple’s settings (see the Privacy policy).
The AtmanAI name, brand mark, UI, and original software are owned by the publisher of the app. You retain whatever rights you have in the prompts you type and in outputs to the extent the law and the relevant model license allow. We do not claim ownership of your on-device conversations.
THE APP IS PROVIDED “AS IS” AND “AS AVAILABLE,” WITHOUT WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, AND NON-INFRINGEMENT. We do not warrant uninterrupted or error-free operation, or that outputs will be accurate or appropriate.
TO THE MAXIMUM EXTENT PERMITTED BY LAW, THE PUBLISHER OF ATMANAI WILL NOT BE LIABLE FOR INDIRECT, INCIDENTAL, SPECIAL, CONSEQUENTIAL, OR PUNITIVE DAMAGES, OR FOR LOST PROFITS, DATA, OR DEVICE DAMAGE, ARISING FROM YOUR USE OF THE APP OR OF MODEL OUTPUTS. OUR TOTAL LIABILITY FOR ANY CLAIM RELATING TO THE APP WILL NOT EXCEED THE AMOUNT YOU PAID US FOR THE APP IN THE TWELVE MONTHS BEFORE THE CLAIM (currently zero, unless a paid edition is sold later). Some jurisdictions do not allow certain limits; in those places, our liability is limited to the greatest extent allowed.
These terms are governed by the laws of India, excluding conflict-of-law rules, unless a mandatory consumer-protection law in your place of residence says otherwise. Courts in India have jurisdiction, except where consumer law gives you the right to sue in your home courts.
If a provision is unenforceable, the rest remains in effect. These terms are the entire agreement for use of AtmanAI, together with the Privacy policy and any store-required EULA.
atmanaiapp@gmail.com · Same support address as the public store listing.