Release Focus: "Chat with your Docs" (RAG)

Users must be able to:

• Log into the system and authenticate via Keycloak to access their personal profile
• Upload one or more documents via a file management UI, which are saved to their profile (only they can access)
  • NOTE: Supported formats include .txt and .pdf files, excluding any image content in PDFs
• Create one or more Assistants, which are saved to their profile (only they can access)
• Choose one or more uploaded documents for an Assistant to use via an Assistant management UI
• Select an Assistant to interact with via the Chat UI
• Receive context-aware responses when chatting via RAG with the Assistant's selected set of files
• Chat history is saved to the user profile and can be accessed at a later time; uploaded files and Assistants are also saved

MUST HAVE (Non-negotiable product needs that are mandatory for the team)

• Implement LeapfrogAI API functionality for file, assistants, vector-store, threads, messages, and run endpoints
• UI utilizes threads
• UI implements the ability to create and execute a run with assistants
• UI switches to using assistants API endpoints for CRUD operations
  • Workaround currently in-place using Sveltekit backend integration with Supabase

SHOULD HAVE (Important initiatives that are not vital, but add significant value)

• Integrate auth into LeapfrogAI API (authenticates users for requests to Supabase; enables row-level security)
• UI implements all file management features through the file API endpoints
  • Potential workaround if file not ready is to integrate with Supabase directly, not using the LeapfrogAI API
• Refactor Supabase deployment to use Istio for ingress instead of kong
• Database migration in place to confidently maintain data when upgrading to future versions of LeapfrogAI
• NOTE: If it unblocks front-end progress with minimal rework, consider using OpenAI API while implementing the remaining parts of LeapfrogAI

COULD HAVE (Nice to have initiatives that will have a small impact if left out)

• Monitor context-length of threads and implement 'rolling context' to stay within model token limit.
  • Implement token counting endpoint.

WILL NOT HAVE (Initiatives that are not priority for this specific time frame)

• UI implements "easy buttons" for transcription, translation, and summarization
  • Also requires extending file uploader to support audio file formats
• Implement RAG and model evaluations
• Implement translation capability (requires implementing translations endpoint, UI modification, model changes)