Privacy Policy — AutoDroid Studio
Last updated: 1 October 2026
Last updated: 1 October 2026
AutoDroid Studio runs entirely on your phone. It is a server, not a client: the models run on your device, the data stays on your device, and the developer operates no backend of any kind.
The developer receives no data from this app. There is no analytics SDK, no crash reporting service, no advertising, and no account with the developer. Nothing in the app reports usage, prompts, generated content, or device information to us — because there is no "us" for it to report to.
One third-party component does report to its own vendor: Google's ML Kit, which powers the optional OCR (text-in-image) and barcode-scanning workflow tasks, sends diagnostic data to Google. It is the only such component in the app, it has no opt-out, and it is described in full under "Data collected by third-party components" below.
The app does make network connections, but only to services you point it at or switch on: a model repository you download from, an AI provider you configure, a mail server you enter credentials for, a site template you install. Those are described in full below.
This policy covers the Android application AutoDroid Studio (com.assassinshadow.aiserver), distributed on Google Play.
Contact: junayed.rahman96@gmail.com
AutoDroid Studio turns an Android phone into a local AI inference server. It runs language, speech, image and embedding models on the device's CPU, GPU or NPU, and exposes them over your local network through an OpenAI-compatible HTTP API, an MCP server, a web dashboard, and an in-app chat interface. It also includes a workflow automation system, hostable mini-sites, file-sharing servers, the option to pool several phones into one server, and optional remote access.
All of the following is written to the app's private storage on your phone. None of it is transmitted to the developer. Uninstalling the app deletes all of it.
Chat conversations — only if you turn on chat history; off by default.
Prompt and request logs — the retention period is configurable in Settings.
Tool-call log — only if you turn it on; off by default.
Workflow definitions, run history and variables — when you create workflows.
Secrets (API keys, passwords, tokens, bot tokens, SSH keys) — when you enter them, encrypted via the Android Keystore. If the Keystore is unavailable the app falls back to plaintext storage and says so in the backup panel.
User accounts (usernames and password hashes) — only if you turn on the multi-user system; off by default.
Downloaded models — when you download them.
Knowledge-base documents, memories and embeddings — when you add them.
Mini-site data (for example a shopping list or diary) — when you install and use a site.
Network-share, data-source, remote-access and federation configuration — when you configure it.
SQLite databases you create — when you create them.
Generated text, images, video and audio, and transcripts — when you generate them.
Content reports — when you flag a generated response as offensive or unsafe.
Android's automatic cloud backup is disabled for this app, specifically so that credentials, SSH keys and databases are never copied to Google's servers without your knowledge. The app has its own backup feature instead: it writes to the app's own storage, or to a file share or data source that you have configured and chosen as the destination.
None.
To be explicit, the app contains no Crashlytics, no Google Analytics, no advertising SDK, no advertising ID, and no analytics or attribution SDK that reports to the developer. It does not have a developer backend, so there is no server for your data to be sent to.
Google ML Kit (text recognition and barcode scanning). The OCR and barcode workflow tasks use Google's ML Kit, delivered through Google Play services. ML Kit reports diagnostic and usage data to Google, and Google documents this at developers.google.com/ml-kit/android-data-disclosure. According to that documentation it includes:
device information (manufacturer, model, OS version and build)
the app's package name and version
a per-installation identifier
performance metrics such as latency, API configuration, input and output sizes, feature versions, event types and error codes
It does not include the images you scan or the text or barcode contents recognized from them — recognition itself happens on your device. Google provides no way to switch this reporting off, which is why it is called out here rather than buried: it is the single exception to "nothing leaves your device except to services you chose".
This is disclosed in the app's Google Play Data safety section. If you never use the OCR or barcode workflow tasks, ML Kit is never invoked.
Your text-to-speech engine. The text-to-speech workflow task uses whichever text-to-speech engine is installed and selected in your Android system settings. The app passes the text to that engine and nothing else. Some engines can use online voices; if yours does, that is governed by the engine vendor's privacy policy, not this one.
Beyond these, the only third-party code in the app that talks to a network is code you point at a destination yourself — the services listed below.
Every connection below happens only in response to something you do or configure. Most of them never happen at all unless you use the relevant feature.
huggingface.co — when you browse or download language, image, speech or embedding models. Downloads are anonymous unless you supply your own access token.
Qualcomm AI Hub public assets (qaihub-public-assets.s3.us-west-2.amazonaws.com) — when you download the NPU build of the Whisper speech-recognition model.
Google Play services — the text-recognition and barcode models are delivered by Play services on first use, rather than bundled in the app. ML Kit also reports diagnostic data to Google while it runs; see "Data collected by third-party components" above.
Downloads send only what any download sends (your IP address and a standard request) to the host serving the file. No prompts or personal data are included.
External AI providers — if you configure a remote OpenAI-compatible endpoint (for example api.openai.com), or a remote transcription endpoint, your prompts or audio are sent to that provider and are subject to their privacy policy, not this one. Speech transcription runs on the device by default; it goes to a remote endpoint only if you name one.
MCP servers — remote Model Context Protocol servers you add as tool sources.
Search providers — DuckDuckGo, Google Custom Search, Brave Search, Tavily, Serper or a SearXNG instance, when a workflow or agent performs a web search. Your search query is sent to the provider you selected.
Messaging services — Telegram, Discord, Slack and Matrix, when you add a bot token or webhook to a workflow or a mini-site. Messages go to the chats and channels you specify, and incoming messages are read only from those.
Email servers — SMTP and IMAP hosts you configure for email workflow tasks.
Data sources — SMB, FTP, SFTP, WebDAV, MySQL/MariaDB and PostgreSQL hosts you configure.
RSS feeds, HTTP requests and webhooks — URLs you enter in a workflow.
Some of the built-in site templates call public services on your behalf, only once you have installed that template and only when you use the feature that needs it:
News Digest fetches the RSS feeds in its source list. It starts with a few public news feeds (such as BBC, NPR, The Guardian, Ars Technica and The Verge), which you can edit or remove.
Shopping List can look up products on Open Food Facts and places and shops on OpenStreetMap (Nominatim and Overpass). What you type into that lookup is sent to those services.
Dynamic DNS providers — DuckDNS, Dynu or No-IP, if you enable DDNS. Your public IP address is sent to the provider to keep your hostname current.
SSH tunnel relays — if you enable public Remote Access, the app opens a reverse tunnel through the relay you select (srv.us, Pinggy, or a host of your own).
Tailscale — if you use Tailscale for private remote access. Tailscale is a separate app with its own privacy policy.
Nothing else. The app makes no connection to the internet that is not either listed above or directed at an address you typed in yourself.
While the server is running, the app announces itself on your local network using multicast DNS (Bonjour/Zeroconf), so other devices can find it by name. The announcement includes the server name you chose (reachable as <name>.local), the name of the loaded model, the app version and a random identifier for this installation. It is visible to devices on the same local network, and does not leave it.
Federation (pooling several phones running AutoDroid Studio into one server) is off by default. When you turn it on, the app listens for other copies of the app on your network and shows them to you, but it sends a prompt to another phone only after you have marked that phone as trusted. Requests forwarded to a trusted phone are processed on that phone, which is also under your control.
These are listed separately because they have real privacy weight and none of them is enabled unless you enable it:
Chat history — conversations are not persisted until you opt in.
Tool-call log — not recorded until you opt in.
Remote access — the local server is reachable only on your own network unless you turn on Remote Access, which requires you to set a password.
Federation — no prompt is forwarded to another phone until you trust it.
Network file shares — folders are shared over your LAN only when you pick them, and only the folders you pick.
Multi-user accounts — the server uses a single shared password until you turn this on.
The app requests network access (to serve and make the connections above), Wi-Fi and network state (to show your server's address and announce it on the local network), notifications (for the persistent server notification and workflow notifications you create), a foreground service and wake lock (to keep the server running), start on boot (only if you enable auto-start), and an optional battery-optimisation exemption, which the app explains before asking and which you can decline.
The app requests no access to your contacts, location, camera, microphone, call log, SMS, calendar, or the notifications of other apps. It has no accessibility service, no all-files access, and does not list your installed apps. Files reach the app only through Android's document picker, one folder or file at a time, and only when you choose them.
By default the app's HTTP server is reachable by other devices on the same local network. This is the point of the app, but it is worth being explicit about: anything you can see in the dashboard, another device on your network can also see, unless you set a password or enable the user-accounts system. If you enable Remote Access, the server becomes reachable from the public internet through the relay you chose, gated by the password you set.
Content generated by models running in this app is produced on your device (or by the external provider or trusted phone you routed it to). The app includes an in-app reporting feature for flagging generated content that is offensive or unsafe; reports are stored locally and you can review or export them. They are not sent anywhere automatically. After filing a report you can choose to email it to the developer: the app opens a pre-filled draft in your own mail app, and it is sent only if you send it. That email is then received at the contact address below and used only to review the reported content.
Models you download are third-party works with their own licences and their own behaviour. The app tells you what licence each model carries, but it does not control what a model produces.
This app is not directed to children under 13 (or the equivalent minimum age in your country), and the developer does not knowingly collect data from anyone — including children. It is a developer and power-user tool.
Secrets are encrypted at rest using the Android Keystore. Passwords are stored as hashes and compared in constant time. The server sets security headers, rate-limits authentication attempts, and restricts cross-origin requests to an allow-list you control. Local network traffic is plain HTTP, because the server is reached at a LAN IP address for which a certificate cannot be issued — treat your local network accordingly. Connections the app makes to internet services use HTTPS wherever the service supports it.
Individual data can be cleared from the relevant screen in the app (chat history, logs, workflows, models, sites, accounts, reports).
Uninstalling the app removes everything it stored, including all models, databases, credentials and generated content.
Backup files you wrote to a share or data source of your own are not touched by uninstalling; delete them there.
User accounts created in this app exist only on your own device. There is no account with the developer to delete, and no server-side copy of anything.
Data you sent to a third-party service you configured (an AI provider, a messaging service, a mail server) is held by that service; contact them to delete it.
If this policy changes, the updated version will be published at the same URL and the date at the top will change. Material changes will also be noted in the app's release notes.
Questions about this policy, data requests, and reports about AI-generated content: junayed.rahman96@gmail.com