AI Auto Application

What if instead of generating application code, the AI was the application.

Here is the AI when told to be a TODO application:

How does it work?

Set the application type by setting the APPLICATION_TYPE environment variable. The default is TODO

When the application starts up, we ask the AI to generate a plausible data model and sample data. We then store this in redis.

Then we pass requests to the AI, allowing it to query redis and return a response template. Tailwind and DaisyUI are included by default to give a consistent way to generate a UI.

Does it work?

Kinda. The UI can morph as you navigate because that isn't saved into the context and it sometimes fails to provide a renderable template.

Setup

0. Take a look at app/config/settings.py and create a .env to setyo your application. Make sure redis is running locally.

1. Create a virtual environment (recommended):

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

2. Install dependencies:

pip install -r requirements.txt

Running the Application

1. Start the web application:

source .env
python -m app.main

3. Open your browser and navigate to http://localhost:8000 (or the configured port)

Future ideas

• manage application data entirely within the context window
• allow generating a DDL and using sqlite
• clone an existing application from its web traffic logs
• send it previous responses for UI consistency.