Privacy Policy
Last updated: 24 August 2026
Your conversations never leave your phone. The AI model runs locally, and no message you type is sent anywhere. The app collects a small amount of anonymous usage analytics, and downloads model files and configuration over the internet.
local LLM (“the app”) is published by Raushan Pandey. This policy explains
exactly what the app does and does not collect. It applies to the Android app
distributed as com.ondevicellm.
What stays on your device
- Your messages and the AI’s replies — stored in a private SQLite database inside the app’s sandbox. They are never uploaded, and no one but you can read them.
- Chat titles — generated by the model running on your phone.
- Downloaded model files — kept in the app’s private storage.
- Your settings — such as the model you picked and whether the local API is enabled.
Uninstalling the app, or clearing its storage in Android Settings, permanently deletes all of the above.
What is collected
The app uses Google Firebase Analytics to record anonymous usage events, so we can see which features are used and where things fail.
| Collected | Example |
|---|---|
| Event names | message_sent, model_loaded, api_service_toggled |
| Model name | SmolLM2 360M Instruct |
| Message length in characters | 42 |
| Durations | model load time, reply time |
| Anonymous app instance ID | assigned by Firebase, reset on reinstall |
| Standard device info | device model, OS version, country, app version |
The text of your messages and replies is never included — only how many characters they contained.
Network connections the app makes
- Model downloads — HTTPS requests to the model host you select (by default, huggingface.co). That host will see your IP address, as with any download.
- Configuration — the app reads a Google Cloud Firestore database at startup to fetch the model catalogue and check whether an update is required. It only reads; it never writes, and sends no personal data.
- Analytics — as described above.
The app makes no other network requests. AI responses are generated entirely on your device — no prompt is ever sent to a server.
The optional local API
If you enable API service in the app’s drawer, the app runs an HTTP
server on port 11434, reachable by other devices on your local network.
- It is off by default and runs only while you enable it.
- It has no authentication. Anything on your network that can reach your phone can send prompts to the model and read the replies.
- It never sends data out to the internet; it only answers requests that reach your phone.
Only enable it on networks you trust.
AI-generated content
Responses are produced by open-weight language models running on your device. They can be inaccurate, misleading, or offensive, and are not reviewed by us. Do not rely on them for medical, legal, financial, or other professional advice.
Data sharing
No data is sold, and none is shared with third parties for advertising. There are no ads and no third-party advertising SDKs. Analytics data is processed by Google as our service provider.
Children
The app is not directed at children under 13. Because the AI models are open-weight and unfiltered, their output may not be suitable for children.
Your choices
- Stop analytics — uninstall the app.
- Delete everything — uninstall, or clear the app’s storage.
- Request deletion of analytics data — email the address below with your device model and approximate dates, and we will submit a deletion request to Google.
Changes
Material changes will be reflected in the “Last updated” date above and, where significant, noted in the app’s release notes.