MageGPT Generates a Wasp To-Do App: A 2023 Demonstration

In this October 2023 demonstration, I try MageGPT, a tool from the team behind Wasp, the sponsor of the video. Wasp is a full-stack JavaScript framework that I compare to Ruby on Rails. MageGPT uses a prompt to generate a project using that framework; the demonstration is a small example, not proof that every generated app will work or be production-ready.

I choose a to-do app from the generator's examples. The prompt asks for a page listing tasks, the ability to create, edit and complete them, and a database where each user can see and edit only their own tasks. I choose a blue brand color, balanced creativity and username-and-password authentication.

The creativity setting is presented as a proxy for model temperature: the interface offers a range from conventional to creative. At the time of this demonstration, I say the generator supports username and password, with other sign-in methods planned. Those are historical product details, not a statement about current support.

I explain that Wasp's central configuration defines pieces such as routes and pages, giving the generator a shared structure for connecting front end and back end. I have seen other generated projects struggle with integration; that is my experience, not a guarantee that Wasp eliminates all such errors.

As files are generated, I say the system uses GPT-4 for upfront planning and GPT-3.5 for code generation. The team also told me it built fixes for common generation errors into the process. In this run, the generator displays the project files and reports about 25,000 tokens used.

To run the output, I install Wasp locally, download and extract the generated ZIP and change into the project directory. The video describes Mac and Windows through WSL as supported environments at the time. Check the current documentation for requirements rather than assuming this 2023 setup still applies.

I run a development database migration and encounter a Node version issue. Using nvm, I select Node 18 for this particular project, rerun the migration, name it and start the app with `wasp start`. The steps and version are from the historical demonstration and may differ for newer releases.

In the local app, I sign up with a username and password, add two tasks and mark them complete. I say separate users have separate tasks. The on-screen exercise shows basic sign-up and task interactions; it does not independently test every authorization or security case.

I then demonstrate deployment to Fly.io. I create or use an account, install and authenticate the Fly CLI, and launch the app through Wasp's Fly deployment command. The video says a credit card was needed and describes testing as free at the time; hosting plans, trial terms and costs can change.

The first deployment fails because the generated project uses SQLite. I change the Wasp configuration to use PostgreSQL for the Fly deployment and retry with an available app name. The second attempt finishes, and I open the deployed to-do app in a browser.

The example shows a prompt-generated application running locally and then on the web, after installing dependencies, resolving a Node version issue and changing the production database. I encourage viewers to try other prompts, while the video establishes only the behavior observed in this small to-do app.