ThinkWell Privacy Policy

ThinkWell is a private AI companion that runs open language models entirely on your device — on Android, and on Windows through a local inference engine. There are no accounts, no analytics, and no server of ours behind it. Your conversations happen between you and your own hardware.

What We Collect

Nothing.

ThinkWell has no sign-up or sign-in and no servers of ours that receive your information. It contains no analytics, no crash reporting, no advertising, and no tracking identifiers of any kind. SeMo Lab never sees your conversations, your attachments, your recordings, or anything else — there is no system on our side that could receive them.

On-Device AI

All inference runs locally. The prompts you write, the images and audio you attach, and the replies you receive are processed on your own device and are never sent to any cloud AI service by the app itself — the one exception being content you deliberately send elsewhere, described under When You Choose to Share below. On Windows, the app talks to its bundled inference engine over a connection that exists only inside your computer (localhost) — chat content crosses that internal connection and goes no further.

Data Stored on Your Device

Your chat history, settings, voice recordings, attached images and audio, downloaded model files, a local speed-benchmark result, and — if you turn it on — the optional Memory feature's stored facts all live in the app's private storage, which the operating system shields from other apps. Memory is off by default and can be cleared at any time.

Nothing is uploaded or synced anywhere, and ThinkWell's data is excluded from Android's cloud backup. Deleting a chat removes it from your history; uninstalling the app removes everything it stores.

Network Connections

ThinkWell's only network use is downloading the AI models you choose: plain, unauthenticated requests for public files from huggingface.co and its content network. If Hugging Face cannot be reached, or is too slow to start sending the file, the download falls back to a mirror of the same public file — first hf-mirror.com, then modelscope.cn, which is operated by Alibaba. These mirrors are used only as a fallback, never first. The requests are identical and carry nothing about you — no identifiers, no personal data — though, as with any website, whichever host serves the file can see your device's IP address. Downloads happen only when you ask: tapping Download on a model, enabling Memory (which fetches its small helper model), or running the benchmark before any model is installed. Nothing downloads at startup, and chatting with an installed model is fully offline.

Links you tap — in Settings, in model citations, or in a model's reply — open in your browser. One edge worth knowing on Android: if a model's reply happens to include a link to a remote image, the rich reply view may load that image from the web while displaying it.

Exports & Backups

Exporting a reply (text, Markdown, Word, or PDF) and backing up your chats are user-initiated and go through your system's share or save dialog — you choose the destination, nothing is uploaded automatically, and we never see it. A chat backup file contains your full conversations in readable form, so store or share it with the same care as any personal document.

The .thinkwell extension is used for two different files: a chat backup, and a shared workflow. A workflow file holds only text — a title, an icon, a prompt and the options chosen for it — and a workflow you import runs locally exactly like the built-in ones. The two file types are labelled internally, so importing the wrong one fails cleanly instead of doing something unexpected.

When You Choose to Share

Nothing leaves your device unless you send it. Export, Copy, Share, Send to another AI, and sharing a workflow all hand the content to another app that you pick — a messaging app, an email client, another AI assistant. From that point it is governed by that app's privacy policy, not ours. ThinkWell has no part in it and keeps no record.

Worth knowing about Send to another AI: when a conversation is long enough to need condensing, the summary is written by a model already downloaded on your device, offline. No cloud service is involved in preparing it — only in whatever you choose to paste it into afterwards.

Permissions

ThinkWell requests internet access, used solely to download the models you select, and microphone access, used only when you record voice input for a speech-capable model — recordings stay on the device like everything else. Attaching images or files uses the system file picker, which needs no special permission.

It does not request your camera, location, or contacts.

Children and Minors

ThinkWell collects no personal data from anyone, of any age, and therefore collects none from minors. The app contains no advertising, no in-app purchases, and no user-to-user communication — conversations are only ever between the user and a model running on their own device.

Because no information ever reaches SeMo Lab, there is nothing for a parent or guardian to request the deletion of. If you have any concern about a minor's use of the app, contact us at info@semo-lab.com.

Your Rights

Privacy laws such as the GDPR and the CCPA give you rights to access, correct, export, and delete the personal data a company holds about you. SeMo Lab holds no personal data about ThinkWell users, so there is nothing for us to retrieve, amend, or erase. We do not sell or share personal information, because we never receive any.

You keep full control of the data on your device: delete individual chats, clear your whole history, clear remembered facts, remove downloaded models, or uninstall the app to erase everything at once. There is no cloud copy to clean up afterwards.

Changes to This Policy

If ThinkWell's privacy practices ever change, this page will be updated and the date above revised. Material changes will be described here rather than quietly amended.

Contact

For privacy questions about ThinkWell, contact: info@semo-lab.com