Local model inference
Choose a compatible model, review its size and device requirements, then download it deliberately. Every reply is written on your device. CuriousLM has no AI server.
Private local AI
Private AI that runs on your device. Your chats never leave it.
The model is downloaded once and answers on your device, so what you ask, write and attach stays with you. No account, no ads, and no signal needed.
Internet is needed for the first visit, model downloads and updates.
The short answer
What does local AI mean?
Local AI performs inference on the device you are using. The model and the software that runs it are downloaded first; after that, a prompt can be processed without sending it to a hosted model API. This differs from a private cloud chatbot, which may protect data in transit or limit retention but still computes the answer on remote servers.
Built for everyday work
More than a local model playground.
CuriousLM combines on-device generation with the working context people expect from a modern assistant.
Documents and photos
Attach PDFs, Word documents, text and code files, or photos. CuriousLM reads them on your device, finds the relevant passages and shows which documents a reply drew on.
Projects and instructions
Keep related chats, files and instructions together, and tell CuriousLM about yourself once so every conversation starts with the right context.
See where the model was unsure
Words the model was unsure about stay faintly marked. Tap one to see the other words that nearly came next, with the odds, and take the answer that way instead.
Encrypted on your device
Chats, files, memories and personal settings are encrypted in local storage from the first launch. Add a fingerprint, face, passkey or passcode lock to be asked before CuriousLM opens.
No account or advertising
Open the app and use it without creating a CuriousLM account. There are no ads. Privacy-limited aggregate usage statistics never include chat or file content and can be turned off in Settings.
Where your words go
Your words never leave your device.
Chats, projects, attached files and downloaded models are stored on your device, in the browser's private site storage or the Android app's private storage. Your chats, files, memories and settings are encrypted there. There is no CuriousLM account and no copy on a server.
- Every reply is written on your device. CuriousLM has no AI server to send your chats to.
- A web search happens only when you ask for one, straight from your device to Wikipedia or DuckDuckGo. CuriousLM has no server in between.
- Models download only when you ask, directly from Hugging Face, and are checked against their published checksum.
- Code a model writes runs in an isolated sandbox with no network access.
- Privacy-limited aggregate analytics never includes chat or file content and can be turned off in Settings.
Choose the right architecture
Local AI and cloud AI solve different problems.
| Question | On-device AI | Hosted cloud AI |
|---|---|---|
| Where is inference performed? | Your phone or browser | A provider's servers |
| Can chat work offline? | Yes, after setup | Usually no |
| What limits model size? | Device memory and storage | Provider infrastructure and plan |
| How current is built-in knowledge? | Limited to the downloaded model | May include newer models and connected tools |
| Who controls the model files? | The user's device | The service provider |
Privacy depends on the complete implementation, not the word “local” alone. Read our detailed local AI vs cloud AI guide.
Models you can inspect
See the cost before you download.
Local models vary significantly in storage, memory use, modality, licence, and speed. CuriousLM shows each model's size, strengths and licence, tells you how well it suits your device, and verifies every download before it is used.
How to choose a model for Android
Common questions
Private local AI, explained plainly.
Does CuriousLM work without the internet?
Yes. After the app and a model have been downloaded, chat works offline. The first visit, model downloads, updates and web search need a connection.
Are prompts sent to a CuriousLM server?
No. Prompts and responses are processed on your device and CuriousLM has no chat server. The app only goes online for updates, the models, voices and Pocket Wikipedia you choose to download, a web search when you ask for one, a report you choose to send, and privacy-limited usage statistics that you can turn off.
Are my chats encrypted?
Yes. Chats, messages, attached files, memories and personal settings are encrypted on your device with a key that never leaves it. In a browser without a lock, that protects the stored files but not someone using your unlocked browser, so you can add a passkey or passcode. The Android app keeps the key in the phone's secure hardware and can ask for your fingerprint. There is no recovery if you forget a passcode.
Does CuriousLM tell me when an answer might be wrong?
Yes, CuriousLM shows where the model was unsure. Words the model picked with low probability stay marked, and you can see the alternatives that were weighed. That is a useful hint, but a model can also be confidently wrong, so check anything important.
Which files can local AI work with?
PDF, Word (DOCX), plain text, Markdown, CSV, HTML, JSON and source code files, plus photos: models that can see look at them, and in the Android app the writing in a photo of a letter or bill is read on the phone, so any model can explain it. Scanned PDFs without a text layer are not read yet.
Will every phone run every local model?
No. Speed depends on memory, graphics support and the model itself. CuriousLM shows each model's size and how well it suits your device before you download it, and recommends one that runs well.
No account required
Try private local AI in your browser.
Open CuriousLM, choose a model that suits your device, and every chat stays with you.
Open CuriousLM