Currently our tool is built on the best of OpenAI's offerings, incorporating all the recent updates announced last week. Our GPTs utilize the Assistant API for an engaging chatbot experience. Additionally, we leverage replicate.com to integrate a variety of open-source models. For instance, we currently offer music generation capabilities (this may be buggy) within our GPTs via replicate and our GPTs fallback to a fine-tuned SD model if the prompts are too rich for DALL-E-3. We are planning to expand our model library rapidly in the near future.

We view the recent OpenAI GPT-builder announcement as a validation of our product - our goal is to use the best of open-source and proprietary models to build these GPTs. Over the past several months, our small but dedicated team (bootstrapped & effectively 2.5 people strong) has been hard at work. We launched our beta version in early August and have since attracted over 1K users.

Here are a few example prompts and GPTs which were created using Appstorm: - ELI5 for STEM topics GPT: https://beta.appstorm.ai/share?url=3321dcba - Research assistant GPT summarises research pdfs: https://beta.appstorm.ai/share?url=86176d7d - Fashion and style advisor GPT: https://beta.appstorm.ai/share?url=5737f2c8 - Dietary label image reader GPT with suggestions: https://beta.appstorm.ai/share?url=0196fdb5 - GPT that provides an absurd and humorous analysis of the latest news as if it were a plot in a sci-fi movie: https://beta.appstorm.ai/share?url=aad29b48 - Superhero Retirement Life GPT: https://beta.appstorm.ai/share?url=fd9689b9 - GPT image generator of well-known historical period reimagined with a twist: https://beta.appstorm.ai/share?url=2aaea9d3

We also have a nifty ideas generation feature where a seed/keyword can spit out relevant ideas, leading to some truly awesome (and sometimes bizarre) creations!

Our GPTs are interactive (prompt must have the 'gpt' keyword), featuring the Assistant API integration and Google news support. The apps are focused on single-function tasks but offer various input and output modalities, including text, audio, and images.

We're also looking ahead: our eventual (and slightly more difficult) goal is to enable users to download these apps (including the Gradio source code) and run them on personal devices or servers using open-source models. This approach might incur a fidelity-loss but will provide more flexibility for enthusiasts and tinkerers. We believe our value is to map the experience of a GPT built with OpenAI models to one that runs equally well with open-source models.

We're thrilled about Appstorm and would love for the Hacker News community to try it out and share your thoughts. We welcome all kinds of feedback – the good, the bad, and the ugly – as it's all incredibly valuable to us!

Thanks for reading! You can find us here: https://www.appstorm.ai