A personal local AI agent that lives on your phone
A personal local AI agent is an AI assistant that runs on your own device, keeps your data there and takes actions for you, such as saving tasks or scheduling reminders, without sending requests to a cloud model. Corenyth is a personal local AI agent for Android: a fine-tuned Qwen3-0.6B model runs on your phone through llama.cpp, and the app validates every proposed action before it changes anything.
What is a personal local AI agent?
An AI agent does more than chat. It interprets a request, decides on an action and carries it out: creating a task, setting a reminder, remembering a preference or asking a clarifying question.
A local agent does all of this on your device. The language model, your conversation history and your personal data never leave the phone. A personal agent also remembers context about you, so later requests can build on earlier ones.
Local AI agent vs cloud AI assistant
| Personal local AI agent (Corenyth) | Typical cloud AI assistant | |
|---|---|---|
| Where the model runs | On your Android phone | On remote servers |
| Internet required | No. The release app has no INTERNET permission | Yes, for most requests |
| Where your data lives | Local SQLite on the phone | Usually in a cloud account |
| Account needed | No | Usually |
| Model size | Small (Qwen3-0.6B, about 364 MiB) | Large, server-hosted models |
| Language ability | Narrower; accuracy still under evaluation | Broader general knowledge |
What Corenyth's local agent can do today
- Create, find, complete and postpone tasks from typed chat, for example “Remind me to call Kabir at 6pm”.
- Schedule Android reminders that fire while the app is closed, including snoozes and daily schedules.
- Remember personal facts and preferences, retrieve them with full-text search, and correct or forget them on request.
- Ask “Which one?” or missing-time questions that survive chat switches and restarts.
- Send optional preparation nudges, meeting check-ins and an evening review.
- Work from two home-screen widgets: a 4 × 2 task list and a 2 × 2 animated companion with a quick chat panel.
How the agent keeps actions safe
The model proposes; the app decides. Corenyth asks the local model for a grammar-constrained, structured plan. App code then checks that plan against your words, task identity and exact dates before saving data or scheduling alarms.
Ambiguous requests trigger a clarifying question. If the model is missing or a plan is rejected, built-in rules handle supported requests. There is no cloud fallback.
Current limits
- Typed chat only: voice input and output are not integrated.
- Android only: there is no iOS, desktop or web app.
- Broad language reliability is not yet accepted, and performance varies by device.
- Only Corenyth's own fine-tuned model file is accepted; other GGUF models are not supported.
Try Corenyth
Download the Corenyth Android app (APK) and the fine-tuned model from the public GitHub release, place the model in Documents/localmodels, and start typing requests. Setup takes a few minutes and needs no account.
Frequently asked questions
Can an AI agent run fully locally on a phone?
Yes. Small language models can run on modern phones. Corenyth runs a fine-tuned Qwen3-0.6B model on 64-bit ARM Android phones through llama.cpp, with no cloud model and no internet permission in the release app.
Is a local AI agent more private than a cloud assistant?
A local agent keeps your requests and personal data on your device, so they are not processed by a remote server. Corenyth stores conversations, tasks, reminders and memories in local SQLite and needs no cloud account.
What can a personal local AI agent not do?
A small on-device model has less general knowledge than large cloud models and cannot browse the web. Corenyth focuses on tasks, reminders, personal memory and follow-through rather than open-ended research.