snseam/awesome-agent-memory
Developer tools이 페이지는 AI로 자동 번역되었습니다. 기술적인 세부 사항은 원문을 확인하세요. 원문 보기
Curated reading list & living survey on long-term memory for LLM agents — 989 papers, 529 archived PDFs, deep notes mapped to memory-kernel modules. Feeds the Ymem Research Radar.
개요
awesome-agent-memory is not only an awesome list. It is a structured evidence base for deciding how long-term memory should be designed, evaluated, and operated in LLM agent systems. The repository separates five evidence layers: The core maintenance rule is simple: keep direct evidence, maintainer inference, vendor self-claims, affiliated evaluations, and independent reproductions in separate buckets. Start with the path that matches your question: Users of skill-enabled agent tools can also invoke the repo-local skill $awesome-agent-memory to use this repository as an evidence-backed review tool. Prerequisite: open this repository in an agent tool that loads .codex/skills/*/SKILL.md, or install or enable this skill in your agent environment. If $awesome-agent-memory is not available, use .codex/skills/awesome-agent-memory/SKILL.md as the instruction and pass this checkout as AAM_ROOT.
README
원문 README의 일부입니다. 원문 보기.
Table Of Contents
- What This Repository Is
- At A Glance
- How To Use It
- Evidence Model
- Repository Map
- Repository Layout
- Scope Boundaries
- Issue And PR Principles
- External Content And Agent Safety
- License And Archival Policy
- Origin And Maintenance
- Contributing
What This Repository Is
awesome-agent-memory helps researchers and builders trace memory-design
decisions to papers, documented product interfaces and evaluation protocols.
Use it to compare approaches, find gaps and design experiments. Inclusion records
relevance and evidence; it is not a product recommendation, security certification
or leaderboard position.
The repository separates five evidence layers:
| Layer | What it owns | Main entry point |
|---|---|---|
| Papers | Scholarly claims, methods, and local reading notes | papers/index.md |
| Products | Public behavior of memory products and memory-enabled platforms | docs/products-landscape.md |
| Benchmarks | Evaluation protocols, usage events, vendor claims, and critique records | docs/benchmarks-landscape.md |
| Synthesis | Taxonomy, living survey, architecture patterns, source maps, and signal logs | docs/README.md |
| Decision workflow | How evidence becomes ImpactReports, experiments, and ADR inputs | docs/research-radar.md |
The core maintenance rule is simple: keep direct evidence, maintainer inference, vendor self-claims, affiliated evaluations, and independent reproductions distinct. Note status, source version and review limitations matter more than the number of links or the submitter’s identity.
At A Glance
추천 도구
다른 키워드를 입력하거나 필터를 제거해 보세요.
설치
npx skillfish add snseam/awesome-agent-memory