Workspace generation using LLM

[kantord]

In some AI tools such as ChatGPT, there is a way to create your own assistants for specific use cases, which in some ways is similar to our workspaces feature. The custom assistants in ChatGPT are called "GTPs". And to create your custom GPT, you can talk with a "GPT creation assistant" which asks you a set of questions, and based on your answers it configured a custom assistant for you.

I think that our workspaces feature could benefit from a similar tool.

[kantord]

Note: this would likely require us to run our own LLM (either locally through ollama or through a cloud function for instance). Offloading this generation to a provider set up by the end user is feasible in theory, but it would be problematic since it might not be the same model that the feature was designed and tested with.

Ollama can also be problematic, since if the user is already running a model or two locally, running a decent model for this purpose might cause performance issues.

So likely the solution would be the creation of a cloud function that calls a remote LLM provider for this, but that would also create the problem of preventing abuse, etc etc.