Smriti FAQs

Question 1

What are the key technical components powering Smriti?

Accepted Answer

Smriti is built on LadybugDB, an embedded property graph database. It integrates with OpenAI-compatible LLM and embedding APIs for encoding and retrieval, uses advanced algorithms like Leiden for clustering, and offers flexible backup options (GitHub, S3).

Question 2

What is Smriti and its primary function?

Accepted Answer

Smriti is a Model Context Protocol (MCP) server designed to provide persistent, graph-based memory for Large Language Model (LLM) applications. Its core function is to enhance LLM recall with human-like associative memory capabilities.

Question 3

What is Leiden Community Detection and how does it benefit Smriti?

Accepted Answer

Leiden Community Detection is an automatic clustering algorithm used by Smriti to group related memories within the graph. This enables 'cluster-aware' retrieval, making memory access more efficient and scalable, especially with thousands of memories, by focusing on relevant topic clusters.

Question 4

How does Smriti achieve human-like memory recall for LLMs?

Accepted Answer

Smriti utilizes EcphoryRAG-inspired multi-stage retrieval. This process combines cue extraction, graph traversal, vector similarity search, and multi-hop associations to deliver highly relevant and associatively linked memories to the LLM.

Question 5

Can Smriti handle multiple users and manage memory over time?

Accepted Answer

Yes, Smriti supports multi-user environments, scaling to thousands of isolated memory stores using separate LadybugDB instances. It also features automatic consolidation, which includes exponential decay, pruning of weak memories, strengthening of frequently accessed ones, and periodic re-clustering to manage memory lifecycle effectively.

Smriti

Smriti

主な機能

ユースケース

主な機能

ユースケース