SME
An open-source long-term memory plugin for AI systems that uses spatial partitioning and dynamic decay mechanisms to manage memory. It supports multi-signal interpretable retrieval, dual-timeline modeling, and Personalized PageRank (PPR) graph expansion. The plugin connects to various AI agents via the Model Context Protocol (MCP).
Timeline (last 90 days)
SME (Spatial Memory Engine) is an AI long-term memory plugin open-sourced by an independent developer, which connects to agents via the MCP protocol and is open-sourced under the Apache 2.0 license.
SME divides memories into different spatial partitions, combined with a dynamic decay mechanism that allows distant, low-value information to fade out while important information is retained and reinforced.
At the retrieval layer, SME adopts a 'near-far fusion' multi-signal strategy, integrating 8 types of signals for interpretable retrieval.
In a blind evaluation battle designed by the author consisting of 40 questions and three groups of random seeds, SME ranked first with a score of 96.7%, with competitors including MEM0, GRAFFITY, and five other mainstream memory solutions.
After replaying the same conversation three times, SME's performance rose from 92.5% to 95%, while the comparison solution LORAC dropped from a high position to 87.5%.
SME's evaluation data all comes from assessments designed by the author, and lacks independent third-party reproduction.
According to the author's disclosure, SME achieves approximately 1.79x improvement in write throughput, 7.4x improvement in batch ingestion, and 127 to 159x faster graph retrieval speed.
SME supports BITEMPRO bi-temporal modeling, which can simultaneously track the time an event occurred and the time the memory was recorded.
SME supports the PPR (Personalized PageRank) graph expansion feature.
SME has a built-in 'idle auto-organization' mechanism that automatically sorts, merges, and archives memories when the system is idle.
1 more timeline events
All Facts (11)
SME is equipped with 190 automated regression tests to safeguard code quality.
50%UnverifiedSME divides memories into different spatial partitions, combined with a dynamic decay mechanism that allows distant, low-value information to fade out while important information is retained and reinforced.
50%UnverifiedAt the retrieval layer, SME adopts a 'near-far fusion' multi-signal strategy, integrating 8 types of signals for interpretable retrieval.
50%UnverifiedIn a blind evaluation battle designed by the author consisting of 40 questions and three groups of random seeds, SME ranked first with a score of 96.7%, with competitors including MEM0, GRAFFITY, and five other mainstream memory solutions.
50%UnverifiedAfter replaying the same conversation three times, SME's performance rose from 92.5% to 95%, while the comparison solution LORAC dropped from a high position to 87.5%.
50%UnverifiedSME's evaluation data all comes from assessments designed by the author, and lacks independent third-party reproduction.
50%UnverifiedAccording to the author's disclosure, SME achieves approximately 1.79x improvement in write throughput, 7.4x improvement in batch ingestion, and 127 to 159x faster graph retrieval speed.
50%UnverifiedSME supports BITEMPRO bi-temporal modeling, which can simultaneously track the time an event occurred and the time the memory was recorded.
50%UnverifiedSME supports the PPR (Personalized PageRank) graph expansion feature.
50%UnverifiedSME has a built-in 'idle auto-organization' mechanism that automatically sorts, merges, and archives memories when the system is idle.
50%UnverifiedSME (Spatial Memory Engine) is an AI long-term memory plugin open-sourced by an independent developer, which connects to agents via the MCP protocol and is open-sourced under the Apache 2.0 license.
50%