Convolution

AI learning app that builds a model of your knowledge and helps you learn what you don't know

Installation

First, for local inference, install ollama, then start it with

OLLAMA_HOST=0.0.0.0 OLLAMA_ORIGINS=* ollama serve

You might have to disable ollama system service first:

sudo systemctl stop ollama
sudo systemctl disable ollama

The reason for doing this is that default ollama doesn't handle CORS well.

Now, you also need to setup a huggingface account at https://huggingface.co/ Then go to https://huggingface.co/settings/keys and add ssh key from ~/.ollama/id_ed25519.pub This will allow you to download gated models like gemma3n. Now, go to gemma3n page at

Now, proceed to download the model. Note that it is 3 GB in size.

ollama run hf.co/unsloth/gemma-3n-E2B-it-GGUF:Q2_K_XL

After this, go ahead and setup this project:

npm install

Now, to run a local app, do

ionic serve

Installation on anroid

First, download the gemma3n model from huggingface at https://huggingface.co/google/gemma-3n-E2B-it-litert-preview

Then copy the file you downloaded to your device:

adb push  gemma-3n-E2B-it-int4.task /data/local/tmp/llm/

Then build the js app, sync it with native android sources, build the android app and copy it on the device. You can conveniently do all that with a single command:

npx cap run android

You need android studio installed for this command to work. Also, it would ask you if to pick your device (or emulator) to which to copy the app.

Development

To create a mobile version, first do

ionic capacitor sync

to build static files and copyt them into native apps.

Then run for example

ionic capacitor run android

to run the app on the connected android phone or an emulator. Please note that you would need android studio installed for this to work.

You can also open android studio project if you wish to modify java part of the code:

ionic capacitor open android