MemPalace Model Card
这是可索引项目报告证据页：它保留 MemPalace Model Card 的源材料入口、机制线索和限制提醒；正文仍需 reader/editor 与 academic public-copy review 后才能当作最终结论引用。
MemPalace Model Card
| Field | Value |
|---|---|
| Repository | MemPalace/mempalace |
| Category | Benchmarked Agent Memory System |
| Stars / forks snapshot | 52800 / 7000 |
| Language | Python |
| License | MIT |
| Raw capture | raw-github/mempalace_mempalace.md |
| Updated by | hourly public metadata update, 2026-05-25 |
1. Role in Self Evolve
MemPalace 是 local-first agent memory system，强调 verbatim storage、可插拔检索、知识图谱和 MCP tools，并公开 LongMemEval/LoCoMo/ConvoMem/MemBench 复现实验说明。
2. Working Principle
conversation/project text -> palace index -> scoped semantic retrieval + graph timeline -> agent wake-up context
3. Evidence Path
web GitHub page observed MIT license, Python project, 52.8k stars and 7k forks; README claims verbatim local memory, ChromaDB-backed retrieval, knowledge graph, MCP tools, and reproducible benchmark files with 96.6% LongMemEval raw R@5 plus 98.4% held-out hybrid R@5. Treat benchmark numbers as project claims until independently rerun. Shell GitHub API access remained blocked by DNS and local gh auth was invalid, so this card treats the current snapshot as web-observed rather than API-verified.
4. Teaching Use
Use this card to explain Benchmarked Agent Memory System in the raw -> classification -> project card -> site/report pipeline. The key question is whether memory/context evidence improves downstream agent behavior, or only improves retrieval on a local benchmark.
5. Limits
当前未克隆源码，未运行 benchmark、SDK examples、skill install flows、memory experiments、agent evolution loops 或 production deployments；star/fork/commit 快照来自公开 GitHub 页面文本或可见页面片段。