Evilander commited on
Commit
ae6cc1e
·
verified ·
1 Parent(s): 15f9cd9

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +67 -0
README.md ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ tags:
4
+ - ai-memory
5
+ - ai-agents
6
+ - mcp
7
+ - benchmark
8
+ - retrieval
9
+ pretty_name: Audrey Memory Benchmark Artifacts
10
+ size_categories:
11
+ - n<1K
12
+ ---
13
+
14
+ # Audrey Memory Benchmark Artifacts
15
+
16
+ This dataset publishes Audrey's local benchmark artifacts for inspection and reproducibility. Audrey is a local-first memory runtime and MCP server for AI agents.
17
+
18
+ Project: https://github.com/Evilander/Audrey
19
+ Package: https://www.npmjs.com/package/audrey
20
+ AMB submission: https://github.com/vectorize-io/agent-memory-benchmark/issues/11
21
+
22
+ ## What is included
23
+
24
+ - `summary.json`: local memory regression summary
25
+ - `report.html`: static benchmark report
26
+ - `local-overall.svg`: overall score graph
27
+ - `retrieval-overall.svg`: retrieval suite graph
28
+ - `operations-overall.svg`: operations suite graph
29
+ - `published-locomo.svg`: reference comparison graph
30
+ - `audrey-benchmark-report.png`: browser-rendered full-page screenshot
31
+ - `perf-0.22.2.json`: latency snapshot
32
+
33
+ ## Methodology
34
+
35
+ The local suite uses mock embeddings and a mock LLM to test Audrey's memory behavior deterministically against weak baselines. It is a release regression gate and evidence artifact, not an official AMB, LongMemEval, or LoCoMo leaderboard submission.
36
+
37
+ ## Current local result
38
+
39
+ Generated: `2026-05-01T03:20:07.968Z`
40
+
41
+ Command:
42
+
43
+ ```bash
44
+ node benchmarks/run.js --provider mock --dimensions 64
45
+ ```
46
+
47
+ Audrey scored `100.0%` with `100.0%` pass rate across the local regression suite.
48
+
49
+ ## Performance snapshot
50
+
51
+ Generated: `2026-05-01T03:39:02.431Z`
52
+
53
+ Hardware: Windows x64, Node 25.5.0, AMD Ryzen 9 7900X3D, 62.9 GB RAM.
54
+
55
+ Methodology: mock 64-dim embeddings, mock LLM, SQLite/sqlite-vec, hybrid recall, limit 5.
56
+
57
+ | Corpus size | Encode p50 | Encode p95 | Recall p50 | Recall p95 | Recall p99 |
58
+ |---|---:|---:|---:|---:|---:|
59
+ | 100 | 0.34 ms | 0.54 ms | 0.56 ms | 1.2 ms | 3.5 ms |
60
+ | 1,000 | 0.31 ms | 0.63 ms | 0.65 ms | 1.2 ms | 7.7 ms |
61
+ | 5,000 | 0.28 ms | 0.51 ms | 2.1 ms | 3.6 ms | 23.3 ms |
62
+
63
+ ## Claim boundary
64
+
65
+ These artifacts are published for transparency. Official third-party leaderboard claims require running the relevant third-party harness and publishing those outputs separately.
66
+
67
+ The next public benchmark target is AMB. Audrey has been submitted as an AMB provider candidate in issue #11.