Skip to content

Latest commit

 

History

1 Commit

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Memory-Centric RAG

There is no best RAG solution, only the best fit for your scenario. Building a second brain with AI requires a fundamentally different paradigm.

What Is Memory-Centric RAG?

Traditional RAG is Search-centric: documents go in → index is built → queries search the index. This works for static knowledge bases — company policies, regulatory archives.

Memory-centric RAG is fundamentally different: knowledge continuously grows → real-time memory → natural access. This is designed for dynamic knowledge management — personal notes, second brains, write-and-query-simultaneously workflows.

Two Paradigms

Search-centric RAG Memory-centric RAG
View of knowledge Static document collection Continuously growing memory
Data freshness Depends on index rebuild cycles Real-time; written = remembered
Query mode Active search Natural conversation
Creation & retrieval Separate workflows Intertwined — editing IS memory update
Typical use case Enterprise search, policy Q&A Personal knowledge management, second brain

Core Principles

  1. No pre-built indexes — the retrieval system shares live data with the editor. Content changes are reflected in the very next query.
  2. Minimal context transmission — only compressed matching fragments are sent transiently. Full documents never leave the device.
  3. Multi-dimensional scoring — not just semantic similarity. Finds the most "useful" context, not just the most "similar."
  4. Editing as memory update — every modification enters retrieval scope immediately. No rebuild cycles.

Reference Implementation

  • NoteRich — A browser-based note-taking app with Memory-centric RAG built in. Open a browser, write notes, import files, ask questions. Zero configuration.

Articles

Architecture

See docs/architecture.md for a detailed comparison between Search-centric and Memory-centric RAG architectures, including boundaries, tradeoffs, and deployment patterns.

License

MIT — implementations of this concept are free to use, modify, and distribute. Contributions and discussions are welcome.

About

Memory-Centric RAG - A different paradigm for building a second brain with AI

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors