Enhanced File Management: AI-Driven Descriptions, Categories, and Tags

[robsturgill]

Feature Description

The proposed feature involves using AI models to assist in generating descriptions, providing category suggestions, and offering tag recommendations for files. There are two main approaches to consider: running a local LLM model or utilizing cloud-based LLM APIs.

This feature would be accessible through an "Experimental" tab in the settings page.

Local LLM Model

• Pros: Maintains privacy as everything is processed on the device.
• Cons: Can be resource-intensive and may not function well on all devices due to hardware limitations.

Cloud-Based LLM APIs

• Pros: Offers higher quality generation due to access to larger models and easier implementation since we wouldn't need to manage the models ourselves.
• Cons: Involves costs and potential limits set by the service providers.

Workflow (Initial Phase)

1. User Input: The user identifies the desired model and provides a basic input (just a few words) to give the AI some context.
2. LLM Context: The LLM is supplied data available that may include model name, filename, title, image and user's text input as the main context for generation.
3. Generated Content: The LLM returns the generated content for the user to review.
4. User Approval: If the user likes what they see, they can accept the content, which then gets saved to the model's "munchie.json" file.
5. Potential improvement: Additionally, the LLM could be supplied existing category and tag list to help maintain taxonomy while also suggesting "new" items.

Potential Benefits

This feature could significantly enhance the STL file management experience by making it easier for users to organize their files and find relevant information about them. It's definitely worth exploring, especially since there are already existing LLM models that we could potentially integrate into our system.

[robsturgill]

Implemented Google Gemini in v0.9.0 release, provide users some prompt templates or they can use their own and select what gets sent to Gemini.