awesome-agent-memory

Decision-grade evidence base for long-term memory in LLM agents: papers, product notes, benchmark protocols, architecture maps, and a research-to-ADR workflow.

中文 · English

---

Table Of Contents

1. What This Repository Is
2. At A Glance
3. How To Use It
4. Evidence Model
5. Repository Map
6. Repository Layout
7. Scope Boundaries
8. License And Archival Policy
9. Origin And Maintenance
10. Contributing

What This Repository Is

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:

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 in separate buckets.

At A Glance

Area	Current coverage	Entry point	Use it when you need to
Paper index	989 papers	papers/index.md	Search agent-memory papers and discovery trails.
Paper stubs	988 stubs	papers/stubs/	Track papers that are covered but not yet fully read.
Local PDFs	534 files	papers/pdfs/	Re-read sources and audit paper notes.
Full / seed paper notes	7 full + 2 seed	papers/	Use human-read notes for architectural decisions.
Memory product notes	34 notes	products/	Compare memory layers, memory SDKs, managed memory, and memory-enabled agents.
Product page archives	33 snapshots	products/archives/	Audit product claims after source pages change.
Benchmark catalog	12 catalog entries	benchmarks/index.md	Understand memory benchmarks and where claims come from.
Claims ledger	Structured YAML ledger	benchmarks/claims/claims.yaml	Separate vendor claims, paper evaluations, critiques, and reproductions.
Survey and taxonomy	1 living survey + 6 meta-survey records	docs/agent-memory-survey.md · docs/meta-surveys.md	Build a field-level view before choosing an implementation.
Impact reports	Template only for now	impact-reports/README.md	Promote strong evidence into kernel-design recommendations.

How To Use It

Start with the path that matches your question:

Goal	Read these first
Get the field overview	docs/agent-memory-survey.md, then docs/taxonomy.md
Find relevant papers	papers/index.md, then full notes under papers/
Compare memory products	docs/products-landscape.md, docs/product-memory-architectures.md, docs/product-architecture-diagrams.md
Check why a product was included or rejected	docs/product-discovery-log.md
Evaluate benchmark claims	docs/benchmarks-landscape.md, benchmarks/index.md, benchmarks/claims/claims.yaml
Track new releases and source channels	docs/signals.md, docs/information-sources.md
Turn research into a memory-kernel decision	docs/research-radar.md, then impact-reports/README.md
See Ymem-specific bindings	docs/ymem-binding/README.md

Evidence Model

The repository is organized so claims can be traced back to their source type.

Evidence class	Where it belongs	How to read it
Primary paper evidence	Full notes in papers/ and canonical links in papers/index.md	Can support method and benchmark-protocol claims when the note is full.
Paper stubs	papers/stubs/	Discovery coverage only; upgrade before using in an ImpactReport.
Product behavior	Product notes in products/ and archives in products/archives/	Supports "the vendor says/offers X", not independent performance conclusions.
Vendor benchmark claims	benchmarks/claims/claims.yaml	Must stay labeled as vendor or affiliated evidence unless independently reproduced.
Independent reproductions	benchmarks/claims/claims.yaml	Require enough third-party setup detail to compare against the original claim.
Maintainer synthesis	docs/	Useful for prioritization and design judgment; should not be confused with direct evidence.

Repository Map

flowchart LR
  R["README"] --> D["docs/<br/>survey, taxonomy, sources, signals"]
  R --> P["papers/<br/>index, full notes, stubs, PDFs"]
  R --> PR["products/<br/>notes and page archives"]
  R --> B["benchmarks/<br/>protocol notes and claims ledger"]
  D --> A["architecture maps<br/>product patterns and diagrams"]
  P --> IR["impact-reports/<br/>decision template"]
  PR --> IR
  B --> IR
  IR --> ADR["kernel decisions<br/>experiments and ADRs outside this repo"]
  D -. "optional project binding" .-> Y["docs/ymem-binding/"]

Loading

Repository Layout

Concept And Synthesis Docs

Path	Purpose
docs/README.md	Documentation map and recommended reading order.
docs/agent-memory-survey.md	Living survey of memory architectures, retrieval, consolidation, forgetting, and evaluation.
docs/taxonomy.md	Shared vocabulary for classifying agent-memory systems and memory-kernel responsibilities.
docs/meta-surveys.md	External meta-survey index from late 2025 through 2026 H1.
docs/research-radar.md	Workflow for turning papers, products, and benchmark evidence into ImpactReports and ADR inputs.
docs/information-sources.md	Source catalog for papers, products, communities, and zh-CN information channels.
docs/related-work.md	Discovery-input attribution and scrape provenance.
docs/signals.md	Reverse-chronological release, comparison, and blog signal log.

Products And Architectures

Path	Purpose
products/	34 memory product notes, including Mem0, Letta, Zep, Graphiti, EverOS, MemOS, Redis Agent Memory Server, Supermemory, TencentDB Agent Memory, and platform-managed memory offerings.
products/archives/	33 markdown snapshots of canonical memory product pages.
docs/products-landscape.md	Product landscape by domain and audience.
docs/product-discovery-log.md	Multi-agent product discovery log with Tier A, Tier B, reject, and alias decisions.
docs/product-memory-architectures.md	Cross-product architecture patterns: memory OS, graph/temporal memory, MCP/local-first memory, managed cloud memory, and personal memory.
docs/product-architecture-diagrams.md	Per-product Mermaid diagrams based on public product patterns.

Papers, Benchmarks, And Decision Workflow

Path	Purpose
papers/	7 full paper notes, 2 seed notes, and the master index.md.
papers/stubs/	988 generated stubs for papers not yet fully read.
papers/pdfs/	534 archived PDFs, about 1.8 GB. See the archival policy below.
papers/_scrape/	Reproducibility artifacts: scrape script and dedup JSON.
benchmarks/	Benchmark catalog, protocol notes, and the note template.
benchmarks/claims/	Usage-event ledger for benchmark mentions, vendor claims, critiques, and reproductions.
benchmarks/archives/	Optional source-page snapshots for benchmark pages, repositories, or dataset cards.
docs/benchmarks-landscape.md	Benchmark landscape by capability, usage type, and evidence independence.
impact-reports/README.md	Template for promoting strong evidence into architecture recommendations.

Governance And Project Binding

Path	Purpose
CONTRIBUTING.md	Contribution rules for papers, products, benchmarks, and evidence ledgers.
CODE_OF_CONDUCT.md	Community conduct policy.
SECURITY.md	Security reporting guidance.
CITATION.cff	Citation metadata.
docs/ymem-binding/	Project-specific bindings for the maintainer's Ymem kernel. Safe to skip if you only need the generic evidence base.

Scope Boundaries

Included:

• long-term and multi-session memory for LLM agents;
• memory writing, retrieval, consolidation, forgetting, personalization, provenance, governance, and auditability;
• products that expose memory as a first-class capability;
• benchmarks used to evaluate memory behavior, personalization, temporal reasoning, forgetting, or memory-layer trade-offs.

Excluded or kept only as adjacent context:

• pure vector databases with no memory lifecycle;
• plain RAG middleware that does not model update, consolidation, or forgetting;
• general agent frameworks where memory is not a first-class surface;
• long-context inference or prompt-cache systems by themselves;
• vendor performance claims presented as independent evidence.

License And Archival Policy

Notes and survey content are released under the Apache License 2.0. Quoted excerpts from external papers and articles remain the property of their authors and are used for commentary and research.

Locally archived PDFs in papers/pdfs/ come from sources that permit redistribution, such as arXiv, ACL Anthology, and open OpenReview submissions. If a PDF source restricts redistribution, open an issue and it will be removed. The canonical URL in the corresponding note remains the authoritative source.

Product and benchmark page snapshots are audit backups. They are not republished commercial material. Cite the original URL from the snapshot header for external or commercial use.

Origin And Maintenance

This repo was started by the Ymem project, but the public entry points are intended to remain useful for any agent-memory kernel. Ymem-specific module names, maintainer stance, and internal benchmark choices live under docs/ymem-binding/.

The paper index also uses public awesome-list repositories as discovery inputs. Attribution and scrape artifacts live in docs/related-work.md and papers/_scrape/. Those inputs are discovery sources; notes, synthesis, and maintainer judgments are maintained here.

Contributing

See CONTRIBUTING.md and CODE_OF_CONDUCT.md. In short:

• New paper: add or upgrade a note under papers/, and use a full note before citing it in an ImpactReport.
• New product: add a note under products/, archive the source page when appropriate, and update the product landscape if it is core.
• New benchmark: add or update a benchmark note, then record usage events in benchmarks/claims/claims.yaml.
• New product or benchmark claim: label the source as vendor, affiliated, critique, or independent reproduction.
• New information source: update docs/information-sources.md or docs/related-work.md.

---

Notes are authored by the maintainer. Some stubs and first drafts were accelerated with LLM tooling; full notes should be grounded in the actual PDF, product page, benchmark source, or archived snapshot rather than unverified secondary summaries.

Survey · Docs map · Papers index · Benchmarks · Memory products · Product architectures · Signals · 中文版