[Feature Request]: Global Knowledge Base / Local RAG support

[gaBa-we]

Description

Please consider adding a global library where users can store JSON, TXT, or Markdown files. The AI should be able to reference these files across all chats (Local RAG) so we don't have to upload the same data repeatedly for every new conversation.

Proposed Solution

No response

[m13v]

local RAG for chat apps is underserved. been building knowledge extraction from local sources (browser data, documents, bookmarks) and the biggest learning is the indexing pipeline matters more than retrieval. most people dump files and expect magic but chunking strategy makes or breaks quality. for JSON and markdown specifically, preserving document structure during chunking is critical. splitting at arbitrary byte boundaries gives garbage embeddings. chunk at semantic boundaries (headers for markdown, top-level keys for JSON) and overlap by a few sentences. sqlite with FTS5 plus a small local embedding model is lightweight enough to run on-device without hurting chat performance.

[m13v]

our local knowledge extraction pipeline with semantic chunking and sqlite-backed embeddings: https://github.com/m13v/ai-browser-profile/blob/main/ai_browser_profile/embeddings.py