mempalace-scientific-analysis/CITATION.cff at main · web3guru888/mempalace-scientific-analysis · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
cff-version: 1.2.0
message: "If you use this analysis, please cite it as below."
authors:
  - family-names: Dey
    given-names: Robin
    email: robin@openhubresearch.org
    affiliation: OpenHub Research
  - family-names: Viradecha
    given-names: Panyanon
    affiliation: OpenHub Research
title: "Spatial Metaphors for LLM Memory: A Critical Analysis of the MemPalace Architecture"
version: 1.2.0
date-released: "2026-04-19"
url: "https://github.com/web3guru888/mempalace-scientific-analysis"
repository-code: "https://github.com/web3guru888/mempalace-scientific-analysis"
license: CC-BY-4.0
type: article
keywords:
  - AI memory systems
  - method of loci
  - spatial memory
  - vector databases
  - LLM memory
  - retrieval-augmented generation
  - MCP protocol
  - ChromaDB
  - MemPalace
  - LongMemEval
  - verbatim storage
abstract: >-
  We present a comprehensive technical analysis of the MemPalace architecture,
  an open-source AI memory system applying the ancient method of loci spatial
  metaphor to organize long-term memory for large language models. Through
  independent codebase analysis, benchmark replication, and comparison with
  competing systems, we find that MemPalace's headline retrieval performance
  (96.6% Recall@5 on LongMemEval) is attributable primarily to its verbatim
  storage philosophy combined with ChromaDB's default embedding model
  (all-MiniLM-L6-v2), rather than to its spatial organizational metaphor per
  se. The palace hierarchy operates as standard vector database metadata
  filtering. However, MemPalace makes several genuinely novel contributions:
  a contrarian verbatim-first storage philosophy that challenges
  extraction-based competitors, an extremely low wake-up cost (~170 tokens),
  a fully deterministic zero-LLM write path, and the first systematic
  application of spatial memory metaphors as an organizing principle for AI
  memory systems. We situate these contributions within the broader landscape
  of AI memory architectures, cognitive science research on hierarchical
  memory, and the emerging MCP protocol ecosystem.
preferred-citation:
  type: unpublished
  authors:
    - family-names: Dey
      given-names: Robin
      email: robin@openhubresearch.org
      affiliation: OpenHub Research
    - family-names: Viradecha
      given-names: Panyanon
      affiliation: OpenHub Research
  title: "Spatial Metaphors for LLM Memory: A Critical Analysis of the MemPalace Architecture"
  year: 2026
  url: "https://github.com/web3guru888/mempalace-scientific-analysis"
  notes: "Preprint, submitted to arXiv cs.AI"