{"family":"memory-research","name":"Memory Research & Benchmarks","count":264,"license":"CC-BY-4.0","systems":[{"id":"a-mem--gh-wujiangxu-a-mem","name":"A-MEM","url":"https://github.com/WujiangXu/A-mem","section":"Research / specialised systems","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Atomic-note / Zettelkasten-style","license":"MIT","latest_release":"no releases","deployment":"searched not found","pricing":"searched not found","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"871★ +10/mo Python","storage":"graph","retrieval":"graph-traversal|similarity","persistence":"long-term","update":"append-only","unit":"fact","governance":"inspectable","conflict":"n/a","description":"Treats memories as atomic linkable notes — explicit nod to Zettelkasten knowledge management. Dynamic linking; retroactive memory revision.","cited_cells":24,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/a-mem--gh-wujiangxu-a-mem"},{"id":"ace-agentic-context-engineering--arxiv-2510-04618","name":"ACE — Agentic Context Engineering","url":"https://arxiv.org/abs/2510.04618","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Paper","license":"Apache-2.0 (official code)","latest_release":"last commit 2026-08-24","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/ace-agent/ace (1387★)","storage":"file","retrieval":"injection|similarity","persistence":"cross-session","update":"extraction|consolidation","unit":"skill|fact","governance":"inspectable|plaintext","conflict":"deterministic","description":"Treats context and agent memory as an evolving 'playbook' that a generate→reflect→curate loop updates incrementally, avoiding brevity bias and context collapse. Reports +10.6% on agent benchmarks and matches the top AppWorld production agent. Traction: ICLR 2026; Stanford/SambaNova; popular community implementation (kayba-ai ACE)","cited_cells":12,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/ace-agentic-context-engineering--arxiv-2510-04618"},{"id":"acon--arxiv-2510-00615","name":"ACON","url":"https://arxiv.org/abs/2510.00615","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Context-compression optimisation","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv-cache","retrieval":"attention","persistence":"session","update":"evict-oldest","unit":"kv-token","governance":"inspectable","conflict":"llm-arbitrate","description":"Optimises context compression for long-horizon LLM agents.","cited_cells":23,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/acon--arxiv-2510-00615"},{"id":"agent-kb--arxiv-2507-06229","name":"Agent KB","url":"https://arxiv.org/abs/2507.06229","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Cross-domain experience KB","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"extraction","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Leverages cross-domain experience for agentic problem solving.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/agent-kb--arxiv-2507-06229"},{"id":"agent-reconfigure-reconfigurable-agent--arxiv-2410-21465","name":"AGENT-RECONFIGURE / Reconfigurable Agent","url":"https://arxiv.org/abs/2410.21465","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Skill-library agent memory","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"kv","retrieval":"extraction-pull","persistence":"long-term","update":"accretion","unit":"procedure","governance":"plaintext","conflict":"n/a","description":"Agent that maintains a skill library and reconfigures its toolset per task — procedural memory of successful sub-task solutions.","cited_cells":5,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/agent-reconfigure-reconfigurable-agent--arxiv-2410-21465"},{"id":"agent-s--arxiv-2410-08164","name":"Agent S","url":"https://arxiv.org/abs/2410.08164","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Open agentic computer-use framework","license":"Apache-2.0","latest_release":"v0.3.2 2025-12-16","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"11.1k★ +545/mo Python","storage":"hybrid","retrieval":"similarity|agentic","persistence":"long-term","update":"extraction","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Open agentic framework that uses computers like a human.","cited_cells":23,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/agent-s--arxiv-2410-08164"},{"id":"agent-workflow-memory--openreview-net","name":"Agent Workflow Memory","url":"https://openreview.net/forum?id=NTAhi2JEEE","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Workflow as memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"file","retrieval":"injection","persistence":"cross-session","update":"extraction","unit":"skill","governance":"inspectable","conflict":"llm-arbitrate","description":"Workflow-based memory framework component.","cited_cells":22,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/agent-workflow-memory--openreview-net"},{"id":"agentbench--gh-thudm-agentbench","name":"AgentBench","url":"https://github.com/THUDM/AgentBench","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Benchmark","license":"Apache-2.0","latest_release":"last commit 2026-02-08","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/THUDM/AgentBench (3798★)","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Multi-environment benchmark (ICLR'24) that evaluates LLMs acting as agents in multiple interactive environments, with code and data in the official repo.","cited_cells":12,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/agentbench--gh-thudm-agentbench"},{"id":"agentevolver--arxiv-2511-10395","name":"AgentEvolver","url":"https://arxiv.org/abs/2511.10395","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Efficient self-evolving agent system","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"agent-controlled","unit":"skill","governance":"inspectable","conflict":"llm-arbitrate","description":"Towards an efficient self-evolving agent system.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/agentevolver--arxiv-2511-10395"},{"id":"agentfold--arxiv-2510-24699","name":"AgentFold","url":"https://arxiv.org/abs/2510.24699","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Proactive context management","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv","retrieval":"agentic","persistence":"session","update":"consolidation","unit":"summary","governance":"inspectable","conflict":"llm-arbitrate","description":"Long-horizon web agents with proactive context management.","cited_cells":21,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/agentfold--arxiv-2510-24699"},{"id":"agentic-memory--arxiv-2601-01885","name":"Agentic Memory","url":"https://arxiv.org/abs/2601.01885","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Unified short + long-term management","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity|agentic","persistence":"long-term","update":"consolidation","unit":"episode","governance":"agent-controlled","conflict":"llm-arbitrate","description":"Learning unified long-term and short-term memory management.","cited_cells":20,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/agentic-memory--arxiv-2601-01885"},{"id":"agentic-plan-caching-apc--neurips-cc","name":"Agentic Plan Caching (APC)","url":"https://neurips.cc/virtual/2025/poster/116166","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Test-time plan-template memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv","retrieval":"similarity","persistence":"cross-session","update":"extraction","unit":"skill","governance":"inspectable","conflict":"llm-arbitrate","description":"Extracts, stores, adapts, and reuses structured plan templates from planning stages of agent applications. NeurIPS 2025 poster.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/agentic-plan-caching-apc--neurips-cc"},{"id":"ai-persona--arxiv-2412-13103","name":"AI PERSONA","url":"https://arxiv.org/abs/2412.13103","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Life-long LLM personalisation","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"lifelong","update":"extraction","unit":"profile","governance":"inspectable","conflict":"llm-arbitrate","description":"Life-long personalisation of LLMs.","cited_cells":20,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/ai-persona--arxiv-2412-13103"},{"id":"ait-associative-transformer--arxiv-2309-12862","name":"AiT (Associative Transformer)","url":"https://arxiv.org/abs/2309.12862","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Hopfield + low-rank explicit memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv-cache","retrieval":"attention|similarity","persistence":"session","update":"overwrite","unit":"kv-token","governance":"opaque","conflict":"n/a","description":"Global Workspace Layer combining low-rank explicit memory + bottleneck attention + Hopfield-style associative memory. Decouples memory from the Hopfield network via learnable explicit memory. Outperforms sparse-Transformer baselines on Sort-of-CLEVR. Latest revision Mar 2025.","cited_cells":20,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/ait-associative-transformer--arxiv-2309-12862"},{"id":"alita--arxiv-2505-20286","name":"Alita","url":"https://arxiv.org/abs/2505.20286","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Generalist agentic reasoning","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"file","retrieval":"agentic","persistence":"long-term","update":"agent-controlled","unit":"skill","governance":"inspectable","conflict":"llm-arbitrate","description":"Generalist agent enabling scalable agentic reasoning.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/alita--arxiv-2505-20286"},{"id":"alita-g--arxiv-2510-23601","name":"Alita-G","url":"https://arxiv.org/abs/2510.23601","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Self-evolving generative agent","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"file","retrieval":"agentic","persistence":"long-term","update":"agent-controlled","unit":"skill","governance":"inspectable","conflict":"llm-arbitrate","description":"Self-evolving generative agent for agent generation.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/alita-g--arxiv-2510-23601"},{"id":"alphaedit--arxiv-2410-02355","name":"AlphaEdit","url":"https://arxiv.org/abs/2410.02355","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Null-space knowledge editing","license":"MIT","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"439★ Python","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"fact","governance":"deterministic","conflict":"llm-arbitrate","description":"Null-space-constrained knowledge editing for language models.","cited_cells":25,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/alphaedit--arxiv-2410-02355"},{"id":"androidworld--gh-google-research-android-world","name":"AndroidWorld","url":"https://github.com/google-research/android_world","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Mobile-OS agent benchmark — 116 real Android app tasks","license":"Apache-2.0","latest_release":"","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"1.2k★ Python","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Google DeepMind's AndroidWorld — 116 hand-crafted tasks across 20 real Android apps (Calendar, Files, Markor, Tasks, etc.). Evaluates LLM agents controlling a real Android device via accessibility APIs. Distinct from AndroidEnv (RL-environment) — AndroidWorld is the benchmark suite. Released 2024; SOTA agent (M3A with GPT-4) at ~30% task success.","cited_cells":30,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/androidworld--gh-google-research-android-world"},{"id":"anthropic-circuit-tracing--transformer-circuits-pub","name":"Anthropic Circuit Tracing","url":"https://transformer-circuits.pub/2025/attribution-graphs/methods.html","section":"Recent method papers — theorized, no distinct product","tier":2,"status":"active","last_verified_at":"2026-10-28","type":"Mechanistic interpretability of parametric storage","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"searched not found","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"read-only","unit":"weight","governance":"auditable","conflict":"n/a","description":"Replaces MLP layers with sparse cross-layer transcoders to produce \"replacement models\" where activations correspond to interpretable features; traces attribution graphs showing which features causally influenced which outputs. Includes case studies on factual recall, revealing how facts are stored and retrieved from weights at the circuit level. Open-source tools run on any open-weights model.","cited_cells":16,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/anthropic-circuit-tracing--transformer-circuits-pub"},{"id":"arigraph--arxiv-2407-04363","name":"AriGraph","url":"https://arxiv.org/abs/2407.04363","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"research-code","last_verified_at":"2026-10-28","type":"CLS dual-layer episodic + semantic KG","license":"MIT","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"166★ Inform 7","storage":"graph","retrieval":"graph-traversal|similarity","persistence":"long-term","update":"extraction","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Episodic observations as timestamped vertices; LLM extracts relational triplets into a semantic knowledge graph; episodic edges connect semantic nodes — structural equivalent of hippocampal index binding cortical representations. Dual-layer world model for interactive text agents.","cited_cells":23,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/arigraph--arxiv-2407-04363"},{"id":"armt-associative-rmt--arxiv-2407-04841","name":"ARMT (Associative RMT)","url":"https://arxiv.org/abs/2407.04841","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-05-14","type":"Hopfield-style energy basins","license":"Apache-2.0","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"64★ Jupyter Notebook","storage":"kv-cache","retrieval":"attention|similarity","persistence":"session","update":"overwrite","unit":"kv-token","governance":"opaque","conflict":"n/a","description":"Associative Recurrent Memory Transformer. Stores tokens in Hopfield-style energy basins for constant-time pattern completion.","cited_cells":27,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/armt-associative-rmt--arxiv-2407-04841"},{"id":"atari-100k--arxiv-1903-00374","name":"Atari 100k","url":"https://arxiv.org/abs/1903.00374","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Sample-efficient RL benchmark — 100k env steps on Atari","license":"GPL-2.0 (ALE)","latest_release":"","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"2.3k★ C++ (ALE / atari-py)","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Sample-efficiency benchmark protocol on Arcade Learning Environment (ALE) — agents limited to 100k environment steps (~2 hours of game-play) before evaluation. Introduced by SimPLe (Kaiser et al. 2019) and adopted by EfficientZero, IRIS, BBF, DreamerV3, Storm. Tests world-model / model-based methods where memory-as-imagination matters more than huge replay buffers.","cited_cells":28,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/atari-100k--arxiv-1903-00374"},{"id":"atlas--arxiv-2505-23735","name":"ATLAS","url":"https://arxiv.org/abs/2505.23735","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-29","type":"Omega rule + polynomial features + Muon","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"attention","persistence":"session","update":"parametric-edit","unit":"kv-token","governance":"inspectable","conflict":"n/a","description":"Long-term memory module that learns to memorise the context at test time. Omega rule (sliding-window update over all past tokens), polynomial feature mappings, Muon optimiser. Generalises to \"DeepTransformers\" family. May 2025.","cited_cells":23,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/atlas--arxiv-2505-23735"},{"id":"babilong--arxiv-2406-10149","name":"BABILong","url":"https://arxiv.org/abs/2406.10149","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Distributed multi-fact reasoning across long docs","license":"Apache-2.0","latest_release":"no releases","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"248★ Jupyter Notebook","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Extends 20 bAbI reasoning tasks (fact chaining, induction, deduction, counting, lists) by scattering relevant sentences across PG-19 book text. Scales to 10M tokens. Tests whether models can chain distributed facts, not merely retrieve a single needle.","cited_cells":29,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/babilong--arxiv-2406-10149"},{"id":"bai-lab-memoryos--gh-bai-lab-memoryos","name":"BAI-LAB MemoryOS","url":"https://github.com/BAI-LAB/MemoryOS","section":"Research / specialised systems","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Hierarchical: short / mid / long-term","license":"Apache-2.0","latest_release":"V1.2 2025-07-18","deployment":"searched not found","pricing":"searched not found","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"1.4k★ +40/mo Python","storage":"hybrid","retrieval":"similarity|hybrid-rerank","persistence":"long-term","update":"consolidation","unit":"summary","governance":"inspectable","conflict":"n/a","description":"Hierarchical \"OS\" with Storage / Updating / Retrieval / Generation modules. Short-term → mid-term via FIFO dialogue-chain; mid-term → long-term via segmented paging.","cited_cells":25,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/bai-lab-memoryos--gh-bai-lab-memoryos"},{"id":"balrog--balrogai-com","name":"BALROG","url":"https://balrogai.com/","section":"Memory benchmarks & evaluation","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"LLM/VLM agent benchmark on 6 challenging game environments","license":"MIT","latest_release":"","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"200★ Python","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"BALROG (Benchmarking Agentic LLM and VLM Reasoning on Games) — 2024 benchmark suite covering NetHack, BabyAI, Crafter, MiniHack, TextWorld, Baba Is AI. Specifically targets long-horizon, partial-observability, exploration-heavy games where memory and planning matter. SOTA GPT-4o and Claude 3.5 Sonnet both score <25% on aggregate. Released by AI Safety Institute UK + Cambridge collaborators.","cited_cells":28,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/balrog--balrogai-com"},{"id":"beam-beyond-a-million-tokens-light--arxiv-2510-27246","name":"BEAM (Beyond a Million Tokens) + LIGHT","url":"https://arxiv.org/abs/2510.27246","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Benchmark","license":"MIT code; CC BY-SA 4.0 data","latest_release":"last commit 2026-08-31","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/mohammadtavakoli78/BEAM (164★)","storage":"none-trivial","retrieval":"none","persistence":"long-term","update":"read-only|append-only","unit":"turn|episode","governance":"n/a","conflict":"out-of-scope","description":"ICLR 2026 benchmark of 100 coherent conversations up to 10M tokens with 2,000 validated probes across ten memory abilities, including contradiction resolution and event ordering. It also proposes the LIGHT episodic/working/scratchpad memory method. Traction: ICLR 2026 paper; Mem0 publishes BEAM explainer and results","cited_cells":11,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/beam-beyond-a-million-tokens-light--arxiv-2510-27246"},{"id":"bench-tau-bench-bench--taubench-com","name":"τ-bench (tau-bench / τ²-bench)","url":"https://taubench.com/","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Benchmark","license":"MIT","latest_release":"last commit 2026-10-29","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/sierra-research/tau2-bench (2354★)","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Sierra's benchmark of whether agents can converse with simulated users, call tools and follow policy in domains such as retail, airline, telecom and banking. τ²-bench adds dual-control environments, and later releases add voice and knowledge tasks.","cited_cells":12,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/bench-tau-bench-bench--taubench-com"},{"id":"berkeley-function-calling-leaderboard-bfcl--gorilla-cs-berkeley-edu","name":"Berkeley Function Calling Leaderboard (BFCL)","url":"https://gorilla.cs.berkeley.edu/leaderboard.html","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Benchmark","license":"Apache-2.0","latest_release":"last commit 2026-03-23","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/ShishirPatil/gorilla (13063★)","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Leaderboard that measures how accurately LLMs call functions (tools). It covers single-turn, multi-turn and agentic settings including web search and memory, plus hallucination and format sensitivity.","cited_cells":12,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/berkeley-function-calling-leaderboard-bfcl--gorilla-cs-berkeley-edu"},{"id":"browsecomp--openai-com","name":"BrowseComp","url":"https://openai.com/index/browsecomp/","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Benchmark","license":"MIT","latest_release":"last commit 2026-04-22","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/openai/simple-evals (4686★)","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"OpenAI benchmark that measures whether browsing agents can find hard-to-locate information that may require visiting tens or hundreds of websites.","cited_cells":12,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/browsecomp--openai-com"},{"id":"browseragent--arxiv-2510-10666","name":"BrowserAgent","url":"https://arxiv.org/abs/2510.10666","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Human-inspired web browsing","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"cross-session","update":"extraction","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Builds web agents with human-inspired web-browsing actions.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/browseragent--arxiv-2510-10666"},{"id":"buffer-of-thoughts","name":"Buffer of Thoughts","url":"","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Thought-augmented reasoning","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv","retrieval":"similarity","persistence":"session","update":"append-only","unit":"chunk","governance":"inspectable","conflict":"llm-arbitrate","description":"Thought-augmented reasoning with LLMs. NeurIPS 2024.","cited_cells":22,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/buffer-of-thoughts"},{"id":"cam--arxiv-2510-05520","name":"CAM","url":"https://arxiv.org/abs/2510.05520","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Constructivist agentic memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"consolidation","unit":"chunk","governance":"inspectable","conflict":"llm-arbitrate","description":"Constructivist view of agentic memory for reading comprehension.","cited_cells":21,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/cam--arxiv-2510-05520"},{"id":"cdmem--aclanthology-org","name":"CDMem","url":"https://aclanthology.org/2025.naacl-industry.80/","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"3-stage encoding + graph indexing","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"graph","retrieval":"graph-traversal|similarity","persistence":"session","update":"extraction","unit":"chunk","governance":"inspectable","conflict":"llm-arbitrate","description":"Context-Dependent Memory framework. Three-stage encoding (expert / short-term / long-term) + graph-structured context-dependent indexing. NAACL 2025 Industry.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/cdmem--aclanthology-org"},{"id":"cl-of-llms-survey--dl-acm-org","name":"CL of LLMs Survey","url":"https://dl.acm.org/doi/10.1145/3735633","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Continual learning of LLMs (survey)","license":"Unlicensed","latest_release":"no releases","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"542★","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"weight","governance":"opaque","conflict":"training-time","description":"Comprehensive survey. ACM Computing Surveys 2025. Anchor reference for the continual-learning-of-LLMs subfield.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/cl-of-llms-survey--dl-acm-org"},{"id":"claude-plays-pokemon-anthropic--twitch-tv","name":"Claude Plays Pokemon (Anthropic)","url":"https://www.twitch.tv/claudeplayspokemon","section":"Memory benchmarks & evaluation","tier":2,"status":"active","last_verified_at":"2026-10-28","type":"Anthropic agent demo — Claude 3.7/4 Sonnet plays Pokemon Red on Twitch","license":"","latest_release":"Twitch livestream ongoing; runs distinct per Claude model version","deployment":"not applicable — research demo","pricing":"not applicable — research demo / community project","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv","retrieval":"extraction-pull","persistence":"session-only","update":"agent-controlled","unit":"episode","governance":"opaque","conflict":"agent","description":"Anthropic's public Twitch livestream of Claude (3.7 Sonnet → Claude 4 Opus) playing Pokemon Red. Agent uses screenshot vision + Game Boy controller actions + a scratchpad memory file the model can edit. Demonstrates long-horizon agentic capability publicly; ran continuously from Feb 2025; viewed by hundreds of thousands cumulatively; key data point for Anthropic's marketing around Claude as a long-horizon agent.","cited_cells":24,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/claude-plays-pokemon-anthropic--twitch-tv"},{"id":"cola--arxiv-2503-09263","name":"COLA","url":"https://arxiv.org/abs/2503.09263","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Multi-agent Windows UI automation","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity|agentic","persistence":"long-term","update":"extraction","unit":"skill","governance":"inspectable","conflict":"llm-arbitrate","description":"Scalable multi-agent framework for Windows UI task automation.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/cola--arxiv-2503-09263"},{"id":"comorag--arxiv-2508-10419","name":"ComoRAG","url":"https://arxiv.org/abs/2508.10419","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Cognitive-inspired narrative RAG","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity|hybrid-rerank","persistence":"long-term","update":"consolidation","unit":"episode","governance":"inspectable","conflict":"n/a","description":"Cognitive-inspired memory-organised RAG for stateful narrative reasoning.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/comorag--arxiv-2508-10419"},{"id":"compressive-transformer--arxiv-1911-05507","name":"Compressive Transformer","url":"https://arxiv.org/abs/1911.05507","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"research-code","last_verified_at":"2026-10-23","type":"Recent full-res + compressed older","license":"MIT","latest_release":"0.4.0 2021-10-04","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"165★ Python","storage":"kv-cache","retrieval":"attention","persistence":"session","update":"consolidation","unit":"kv-token","governance":"opaque","conflict":"n/a","description":"Maintains recent states in full resolution while compressing older memories with learned compression functions. DeepMind.","cited_cells":28,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/compressive-transformer--arxiv-1911-05507"},{"id":"context-engineering--x-com","name":"Context Engineering","url":"https://x.com/karpathy/status/1937902205765607626","section":"Theoretical / informal — ideas without a paper","tier":5,"status":"active","last_verified_at":"2026-10-23","type":"Discipline rename","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"none-trivial","retrieval":"injection","persistence":"ephemeral","update":"n/a","unit":"chunk|turn","governance":"n/a","conflict":"out-of-scope","description":"\"+1 for 'context engineering' over 'prompt engineering' … the delicate art and science of filling the context window with just the right information for the next step.\" The LLM is \"a coworker with anterograde amnesia\" — cannot consolidate or build long-running knowledge once training ends.","cited_cells":12,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/context-engineering--x-com"},{"id":"context-engineering-naming-event--simonwillison-net","name":"Context Engineering — naming event","url":"https://simonwillison.net/2025/Jun/27/context-engineering/","section":"Theoretical / informal — ideas without a paper","tier":5,"status":"active","last_verified_at":"2026-10-23","type":"Discipline crystallisation","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"none-trivial","retrieval":"none","persistence":"n/a","update":"n/a","unit":"n/a","governance":"n/a","conflict":"out-of-scope","description":"Endorses \"context engineering\" as a distinct discipline from prompt engineering — what agents do with their context window (routing, compression, tool output formatting, memory retrieval injection) is engineering, not just prompting.","cited_cells":11,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/context-engineering-naming-event--simonwillison-net"},{"id":"context-expansion-law--huyenchip-com","name":"Context Expansion Law","url":"https://huyenchip.com/2025/01/07/agents.html","section":"Theoretical / informal — ideas without a paper","tier":5,"status":"active","last_verified_at":"2026-10-23","type":"Parkinson's-Law analogue for context","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"none-trivial","retrieval":"injection","persistence":"ephemeral","update":"n/a","unit":"token","governance":"n/a","conflict":"out-of-scope","description":"\"Application context tends to expand to fill the context limits supported by the model.\" Treats agent memory as a first-class unresolved design problem rather than a solved component; explicitly defers a memory deep-dive to a future post.","cited_cells":10,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/context-expansion-law--huyenchip-com"},{"id":"crafter--gh-danijar-crafter","name":"Crafter","url":"https://github.com/danijar/crafter","section":"Memory benchmarks & evaluation","tier":3,"status":"research-code","last_verified_at":"2026-10-28","type":"Minecraft-inspired 2D RL benchmark with 22 achievements","license":"MIT","latest_release":"","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"500★ Python","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Danijar Hafner's 2D Minecraft-inspired benchmark — open-world survival with 22 hierarchical achievements (collect wood → make pickaxe → mine stone → ... → collect diamond). Procedurally-generated 64x64 grid world; tests long-horizon planning, exploration, credit assignment. Lightweight (single Python file) — designed as 'inner-loop Minecraft' for fast iteration. Used as flagship benchmark for DreamerV2/V3.","cited_cells":31,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/crafter--gh-danijar-crafter"},{"id":"creator","name":"CREATOR","url":"","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Tool creation for reasoning","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"cross-session","update":"extraction","unit":"skill","governance":"inspectable","conflict":"llm-arbitrate","description":"Tool creation for disentangling abstract and concrete reasoning. EMNLP Findings.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/creator"},{"id":"d-smart--arxiv-2510-13363","name":"D-SMART","url":"https://arxiv.org/abs/2510.13363","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Dynamic structured memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"graph","retrieval":"graph-traversal|similarity","persistence":"long-term","update":"overwrite","unit":"fact","governance":"inspectable","conflict":"llm-arbitrate","description":"Enhances dialogue consistency via dynamic structured memory and reasoning.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/d-smart--arxiv-2510-13363"},{"id":"darwin-g-del-machine--arxiv-2505-22954","name":"Darwin Gödel Machine","url":"https://arxiv.org/abs/2505.22954","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Open-ended self-improvement","license":"Apache-2.0","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"2.0k★ +20/mo Python","storage":"file","retrieval":"agentic","persistence":"long-term","update":"agent-controlled","unit":"skill","governance":"auditable","conflict":"llm-arbitrate","description":"Open-ended evolution of self-improving agents.","cited_cells":27,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/darwin-g-del-machine--arxiv-2505-22954"},{"id":"deepagent--arxiv-2510-21618","name":"DeepAgent","url":"https://arxiv.org/abs/2510.21618","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"General reasoning with scalable tools","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"agentic","persistence":"session","update":"agent-controlled","unit":"chunk","governance":"inspectable","conflict":"llm-arbitrate","description":"General reasoning agent with scalable toolsets.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/deepagent--arxiv-2510-21618"},{"id":"diamond--arxiv-2405-12399","name":"DIAMOND","url":"https://arxiv.org/abs/2405.12399","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"research-code","last_verified_at":"2026-10-28","type":"Diffusion world model for RL","license":"MIT","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"2.0k★ Python","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"trajectory","governance":"opaque","conflict":"n/a","description":"DIffusion As a Model Of eNvironment Dreams. Trains an RL agent inside a diffusion-based world model — denoising diffusion generates high-fidelity environment rollouts rather than discrete latents. Addresses visual-detail loss in prior discrete-token world models.","cited_cells":26,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/diamond--arxiv-2405-12399"},{"id":"differentiable-search-index-dsi--arxiv-2202-06991","name":"Differentiable Search Index (DSI)","url":"https://arxiv.org/abs/2202.06991","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Corpus encoded into model weights","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"document","governance":"opaque","conflict":"n/a","description":"Encodes entire document corpora directly into Transformer parameters; the model itself becomes the index.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/differentiable-search-index-dsi--arxiv-2202-06991"},{"id":"dreamer-4--arxiv-2509-24527","name":"Dreamer 4","url":"https://arxiv.org/abs/2509.24527","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Scalable agent in fast world model","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"trajectory","governance":"opaque","conflict":"n/a","description":"Hafner / Yan. Solves control tasks by training agents inside a fast, accurate world model.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/dreamer-4--arxiv-2509-24527"},{"id":"dreamerv3--arxiv-2301-04104","name":"DreamerV3","url":"https://arxiv.org/abs/2301.04104","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Model-based RL with imagination","license":"MIT","latest_release":"no releases","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"3.2k★ Python","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"trajectory","governance":"opaque","conflict":"n/a","description":"Hafner et al. General world-model agent. Symlog observation, KL balance, RMSNorm, Block GRU.","cited_cells":25,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/dreamerv3--arxiv-2301-04104"},{"id":"dual-process-agent-dpa--mdpi-com","name":"Dual-Process Agent (DPA)","url":"https://www.mdpi.com/2079-9292/15/6/1232","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-05-14","type":"Kahneman System 1 / System 2","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"consolidation","unit":"summary","governance":"inspectable","conflict":"llm-arbitrate","description":"System 1: fast retrieval path that pulls compressed, relevant memories and conditions generation. System 2: slow reflection pass that evaluates outcomes and writes curated updates back through a conservator gate. No model-weight changes.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/dual-process-agent-dpa--mdpi-com"},{"id":"dynamic-cheatsheet--arxiv-2504-07952","name":"Dynamic Cheatsheet","url":"https://arxiv.org/abs/2504.07952","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Test-time adaptive memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv","retrieval":"injection","persistence":"session","update":"overwrite","unit":"chunk","governance":"inspectable","conflict":"llm-arbitrate","description":"Test-time learning with adaptive memory.","cited_cells":20,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/dynamic-cheatsheet--arxiv-2504-07952"},{"id":"elder","name":"ELDER","url":"","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Mixture-of-LoRA model editing","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"fact","governance":"opaque","conflict":"llm-arbitrate","description":"Enhances lifelong model editing with mixture-of-LoRA. AAAI 2025.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/elder"},{"id":"em-llm--arxiv-2407-09450","name":"EM-LLM","url":"https://arxiv.org/abs/2407.09450","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Bayesian-surprise episodic partition","license":"MIT","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"272★ Python","storage":"kv-cache","retrieval":"similarity|attention","persistence":"long-term","update":"consolidation","unit":"episode","governance":"inspectable","conflict":"n/a","description":"Uses Bayesian surprise detection to partition the cache into meaningful episodes for coherent retrieval.","cited_cells":26,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/em-llm--arxiv-2407-09450"},{"id":"emat--arxiv-2210-16773","name":"EMAT","url":"https://arxiv.org/abs/2210.16773","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"research-code","last_verified_at":"2026-10-28","type":"Sub-ms KV memory for QA pairs","license":"Unlicensed","latest_release":"no releases","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"35★ Python","storage":"kv-cache","retrieval":"similarity","persistence":"long-term","update":"read-only","unit":"kv-token","governance":"opaque","conflict":"n/a","description":"Efficient Memory-Augmented Transformer. Encodes millions of QA pairs into key-value memory; MIPS search returns hits in sub-millisecond latency. EMNLP 2022.","cited_cells":24,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/emat--arxiv-2210-16773"},{"id":"emu-efficient-episodic-memory-utilization--arxiv-2403-01112","name":"EMU (Efficient Episodic Memory Utilization)","url":"https://arxiv.org/abs/2403.01112","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"research-code","last_verified_at":"2026-10-28","type":"Coherent-memory MARL","license":"Apache-2.0","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"56★ Python","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"consolidation","unit":"episode","governance":"opaque","conflict":"llm-arbitrate","description":"Multi-agent RL with semantically coherent episodic buffer; selectively promotes desirable transitions to avoid local convergence.","cited_cells":25,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/emu-efficient-episodic-memory-utilization--arxiv-2403-01112"},{"id":"engram-deepseek--arxiv-2601-07372","name":"Engram (DeepSeek)","url":"https://arxiv.org/abs/2601.07372","section":"Recent method papers — theorized, no distinct product","tier":1,"status":"active","last_verified_at":"2026-10-23","type":"Conditional-memory N-gram lookup module","license":"Apache-2.0 (code); Engram model use under a separate Model License","latest_release":"no releases","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"searched not found","llm_lock":"","github":"deepseek-ai/Engram — ~4.6k stars, 4 commits, last commit not confirmed","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"read-only","unit":"chunk","governance":"opaque","conflict":"n/a","description":"DeepSeek + PKU. Introduces \"conditional memory\" alongside MoE's conditional compute. 27B-param N-gram lookup store offloads static knowledge from the transformer; O(1) lookup; prefetched from host RAM at near-zero runtime overhead.","cited_cells":32,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/engram-deepseek--arxiv-2601-07372"},{"id":"episodic-memory-is-the-missing-piece--arxiv-2502-06975","name":"Episodic Memory Is the Missing Piece","url":"https://arxiv.org/abs/2502.06975","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-23","type":"CLS framework as design requirement","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"n/a","persistence":"long-term|parametric-permanent","update":"consolidation|append-only","unit":"episode","governance":"n/a","conflict":"out-of-scope","description":"Pink et al. — position paper. Uses CLS theory to identify five structural properties of episodic memory absent from current LLM agents: single-shot acquisition, temporal context, source binding, flexible recombination, selective consolidation. Maps each to a research-roadmap requirement.","cited_cells":16,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/episodic-memory-is-the-missing-piece--arxiv-2502-06975"},{"id":"evermemos--arxiv-2601-02163","name":"EverMemOS","url":"https://arxiv.org/abs/2601.02163","section":"Research / specialised systems","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Episodic + semantic + reconstructive","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity|hybrid-rerank","persistence":"long-term","update":"consolidation","unit":"episode","governance":"opaque","conflict":"llm-arbitrate","description":"Self-organizing memory OS for structured long-horizon reasoning. Three-phase model: episodic, semantic, reconstructive.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/evermemos--arxiv-2601-02163"},{"id":"evo-memory--arxiv-2511-20857","name":"Evo-Memory","url":"https://arxiv.org/abs/2511.20857","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Benchmark","license":"unknown","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"","storage":"none-trivial","retrieval":"none","persistence":"cross-session","update":"read-only","unit":"trajectory|episode","governance":"n/a","conflict":"out-of-scope","description":"Google streaming benchmark for test-time learning: agents must search, adapt and update memory across task sequences. It implements 10+ memory modules on 10 datasets and introduces the ExpRAG and ReMem baselines. Traction: Google (Ed Chi, Fernando Pereira et al.) authorship","cited_cells":6,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/evo-memory--arxiv-2511-20857"},{"id":"evolve-mem--openreview-net","name":"EVOLVE-MEM","url":"https://openreview.net/forum?id=dfPQrg1WA5","section":"Research / specialised systems","tier":3,"status":"active","last_verified_at":"2026-05-14","type":"Self-adaptive 3-level hierarchy","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity|hybrid-rerank","persistence":"long-term","update":"consolidation","unit":"summary","governance":"opaque","conflict":"llm-arbitrate","description":"Dynamic Memory Network + Hierarchical Memory Manager + Self-Improvement Engine. L0 raw embeddings, L1 contextual summaries, L2 high-level principles. NeurIPS 2025 (Scaling Environments for Agents workshop).","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/evolve-mem--openreview-net"},{"id":"evomembench--alphaxiv-org","name":"EvoMemBench","url":"https://www.alphaxiv.org/abs/2605.18421","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Benchmark","license":"unspecified","latest_release":"last commit 2026-07-03","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/DSAIL-Memory/EvoMemBench (14★)","storage":"none-trivial","retrieval":"none","persistence":"cross-session|session","update":"read-only","unit":"episode|trajectory","governance":"n/a","conflict":"out-of-scope","description":"Benchmark that crosses memory scope (in-episode vs cross-episode) with content (knowledge vs execution) and compares 15 memory methods with long-context baselines under one protocol. Traction: Standardized comparison of 15 memory methods; HKUST(GZ) and others","cited_cells":11,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/evomembench--alphaxiv-org"},{"id":"ewc-elastic-weight-consolidation--arxiv-1612-00796","name":"EWC (Elastic Weight Consolidation)","url":"https://arxiv.org/abs/1612.00796","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"archived","last_verified_at":"2026-10-23","type":"Quadratic constraint per weight","license":"no-license","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"260★ Jupyter Notebook","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"weight","governance":"opaque","conflict":"n/a","description":"Kirkpatrick et al. Soft quadratic constraint pulls each weight back toward old values, scaled by importance for prior tasks. PNAS 2017.","cited_cells":24,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/ewc-elastic-weight-consolidation--arxiv-1612-00796"},{"id":"expel","name":"ExpeL","url":"","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"LLM agents as experiential learners","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"cross-session","update":"extraction","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"LLM agents are experiential learners. AAAI 2024.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/expel"},{"id":"expel--arxiv-2308-10144","name":"ExpeL","url":"https://arxiv.org/abs/2308.10144","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Experiential learning agent","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"consolidation","unit":"episode","governance":"plaintext","conflict":"n/a","description":"Cross-task generalisation by collecting and abstracting successful trajectories into insights for future tasks.","cited_cells":6,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/expel--arxiv-2308-10144"},{"id":"externalization-in-llm-agents--arxiv-2604-08224","name":"Externalization in LLM Agents","url":"https://arxiv.org/abs/2604.08224","section":"Theoretical / informal — ideas without a paper","tier":4,"status":"active","last_verified_at":"2026-10-23","type":"Memory + skills + protocols as coupled externalization","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"none-trivial","retrieval":"none","persistence":"n/a","update":"n/a","unit":"n/a","governance":"n/a","conflict":"out-of-scope","description":"Traces the shift from weights-as-capability to harness-as-capability; analyzes memory, skills, and protocols as three coupled forms of externalization and examines how they interact. Memory is defined as the externalization of state across time. Covers self-evolving harnesses and shared agent infrastructure as emerging directions. April 9, 2026.","cited_cells":20,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/externalization-in-llm-agents--arxiv-2604-08224"},{"id":"files-are-all-you-need--llamaindex-ai","name":"Files Are All You Need","url":"https://www.llamaindex.ai/blog/files-are-all-you-need","section":"Theoretical / informal — ideas without a paper","tier":5,"status":"active","last_verified_at":"2026-10-28","type":"Filesystem as memory interface","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"file","retrieval":"extraction-pull","persistence":"cross-session","update":"append-only","unit":"file","governance":"inspectable","conflict":"n/a","description":"Coding agents (Claude Code, Cursor) converge on the filesystem as their primary memory abstraction: conversation histories as searchable files, skills as files, retrieval via file search rather than vector DBs. Argument: LLMs are fluent with filesystem concepts, so the filesystem is the right interface even if storage underneath is a database.","cited_cells":11,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/files-are-all-you-need--llamaindex-ai"},{"id":"fincon--arxiv-2407-06567","name":"Fincon","url":"https://arxiv.org/abs/2407.06567","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Conceptual verbal reinforcement","license":"MIT","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"29★ Python","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"extraction","unit":"summary","governance":"inspectable","conflict":"llm-arbitrate","description":"Synthesised LLM multi-agent system with conceptual verbal reinforcement.","cited_cells":24,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/fincon--arxiv-2407-06567"},{"id":"flex--arxiv-2511-06449","name":"FLEX","url":"https://arxiv.org/abs/2511.06449","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Forward learning from experience","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"file","retrieval":"similarity","persistence":"long-term","update":"agent-controlled","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Continuous agent evolution via forward learning from experience.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/flex--arxiv-2511-06449"},{"id":"forever--arxiv-2601-03938","name":"FOREVER","url":"https://arxiv.org/abs/2601.03938","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Forgetting-curve memory replay","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"consolidation","unit":"episode","governance":"opaque","conflict":"llm-arbitrate","description":"Replay scheduler inspired by the Ebbinghaus forgetting curve, plus intensity-aware regularisation to control replay strength. Mitigates catastrophic forgetting in LLM continual learning.","cited_cells":21,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/forever--arxiv-2601-03938"},{"id":"four-type-agent-memory-taxonomy--lilianweng-github-io","name":"Four-Type Agent Memory Taxonomy","url":"https://lilianweng.github.io/posts/2023-06-23-agent/","section":"Theoretical / informal — ideas without a paper","tier":5,"status":"active","last_verified_at":"2026-10-23","type":"Cognitive-science → LLM mapping","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity|attention","persistence":"long-term|ephemeral","update":"n/a","unit":"chunk","governance":"n/a","conflict":"out-of-scope","description":"\"Agent = LLM + memory + planning + tools.\" Maps cognitive memory types to LLM machinery: sensory (learned embeddings), short-term / in-context (the context window), long-term (external vector store with fast retrieval). The first widely-cited taxonomy mapping human memory science onto LLM agent architecture.","cited_cells":11,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/four-type-agent-memory-taxonomy--lilianweng-github-io"},{"id":"from-human-memory-to-ai-memory-survey--arxiv-2504-15965","name":"From Human Memory to AI Memory (survey)","url":"https://arxiv.org/abs/2504.15965","section":"Theoretical / informal — ideas without a paper","tier":4,"status":"active","last_verified_at":"2026-10-23","type":"3D taxonomy: object × form × time","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"none-trivial","retrieval":"none","persistence":"n/a","update":"n/a","unit":"n/a","governance":"n/a","conflict":"out-of-scope","description":"Eight-quadrant classification grid across personal/system, parametric/non-parametric, and short-term/long-term axes. Bridges cognitive-science memory taxonomy to LLM architecture choices, less common than purely engineering-oriented surveys. v2 revision April 23, 2025.","cited_cells":21,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/from-human-memory-to-ai-memory-survey--arxiv-2504-15965"},{"id":"g-memory--arxiv-2506-07398","name":"G-Memory","url":"https://arxiv.org/abs/2506.07398","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Hierarchical multi-agent memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"graph-traversal|similarity","persistence":"long-term","update":"extraction","unit":"summary","governance":"inspectable","conflict":"llm-arbitrate","description":"Tracing hierarchical memory for multi-agent systems.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/g-memory--arxiv-2506-07398"},{"id":"gaia--huggingface-co","name":"GAIA","url":"https://huggingface.co/gaia-benchmark","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Benchmark","license":"searched not found","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Benchmark for general AI assistants made of real-world questions that require reasoning, multimodality, web browsing and tool use, with a public leaderboard on Hugging Face. Traction: ~917 likes on the Hugging Face dataset","cited_cells":5,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/gaia--huggingface-co"},{"id":"gam-general-agentic-memory--arxiv-2511-18423","name":"GAM (General Agentic Memory)","url":"https://arxiv.org/abs/2511.18423","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"JIT runtime research over immutable page-store","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"file","retrieval":"agentic|extraction-pull","persistence":"long-term","update":"read-only","unit":"document","governance":"inspectable","conflict":"n/a","description":"Separates offline memory (lightweight Memorizer highlighting key facts) from online runtime (Researcher agent that queries the full raw page-store at inference). Inverts usual approach — keep full fidelity offline, pay complexity at query time.","cited_cells":21,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/gam-general-agentic-memory--arxiv-2511-18423"},{"id":"generative-agents--arxiv-2304-03442","name":"Generative Agents","url":"https://arxiv.org/abs/2304.03442","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"research-code","last_verified_at":"2026-10-23","type":"Interactive simulacra of human behavior","license":"Apache-2.0","latest_release":"no releases","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"21.2k★","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"append-only","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Park et al. — landmark agent-simulation paper. Reflection + memory stream + retrieval enable believable agent behavior.","cited_cells":27,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/generative-agents--arxiv-2304-03442"},{"id":"generative-agents-park-et-al--arxiv-2304-03442","name":"Generative Agents (Park et al.)","url":"https://arxiv.org/abs/2304.03442","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Memory-stream + reflection + planning","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"accretion|consolidation","unit":"episode","governance":"plaintext","conflict":"n/a","description":"Stanford's iconic memory-stream architecture for believable simulacra — observations + reflections + plans + retrieval by recency/importance/relevance. UIST 2023.","cited_cells":6,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/generative-agents-park-et-al--arxiv-2304-03442"},{"id":"genie--arxiv-2402-15391","name":"Genie","url":"https://arxiv.org/abs/2402.15391","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"research-code","last_verified_at":"2026-10-23","type":"Foundation world model from video","license":"MIT","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"273★ Python","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"read-only","unit":"scene-graph","governance":"opaque","conflict":"n/a","description":"DeepMind. Foundation world model trained unsupervised on internet gaming video. Generates frame-by-frame controllable virtual worlds from text / image / sketch prompts. Spatiotemporal video tokenizer + autoregressive dynamics + latent action model.","cited_cells":24,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/genie--arxiv-2402-15391"},{"id":"h-r--arxiv-2509-12810","name":"H²R","url":"https://arxiv.org/abs/2509.12810","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Hierarchical hindsight reflection","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"cross-session","update":"extraction","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Hierarchical hindsight reflection for multi-task LLM agents.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/h-r--arxiv-2509-12810"},{"id":"h2o","name":"H2O","url":"","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Heavy-hitter oracle KV eviction","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv-cache","retrieval":"attention|cache-lookup","persistence":"session","update":"evict-oldest","unit":"kv-token","governance":"deterministic","conflict":"n/a","description":"Heavy-hitter oracle for efficient generative inference. NeurIPS 2023.","cited_cells":21,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/h2o"},{"id":"h2o-heavy-hitter-oracle--arxiv-2306-14048","name":"H2O (Heavy-Hitter Oracle)","url":"https://arxiv.org/abs/2306.14048","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"KV-cache eviction policy","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"kv","retrieval":"similarity","persistence":"session","update":"replacement","unit":"token","governance":"opaque","conflict":"n/a","description":"Heavy-Hitter eviction policy for KV cache — retains tokens with cumulative high attention scores; up to 5× throughput.","cited_cells":6,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/h2o-heavy-hitter-oracle--arxiv-2306-14048"},{"id":"habitat-3-0-social-agents--aihabitat-org","name":"Habitat 3.0 (social agents)","url":"https://aihabitat.org/habitat3/","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Embodied human-robot social interaction benchmark","license":"MIT (Habitat-Lab); CC-BY-NC (3D scenes)","latest_release":"","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"2.6k★ Python/C++","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Meta AI's Habitat 3.0 — extends Habitat 2.0 with avatar humans, social tasks (navigate-while-following, rearrangement-while-not-blocking), and human-robot collaboration evaluation. Foundational for embodied social-agent research; supports both learned and scripted human avatars; SOTA realistic 3D scanned scenes (HM3D, Replica).","cited_cells":28,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/habitat-3-0-social-agents--aihabitat-org"},{"id":"halumem--arxiv-2511-03506","name":"HaluMem","url":"https://arxiv.org/abs/2511.03506","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Benchmark","license":"CC-BY-NC-ND-4.0","latest_release":"last commit 2026-09-03","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/MemTensor/HaluMem (178★)","storage":"none-trivial","retrieval":"none","persistence":"long-term","update":"read-only","unit":"fact|episode","governance":"n/a","conflict":"out-of-scope","description":"Benchmark that measures hallucination at each stage of an agent memory system (extraction, updating, QA) rather than only final answers; HaluMem-Medium/Long datasets have ~15k memory points and contexts over 1M tokens. Traction: From MemTensor and China Telecom; HF dataset IAAR-Shanghai/HaluMem; widely cited by memory vendors","cited_cells":11,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/halumem--arxiv-2511-03506"},{"id":"hami--nature-com","name":"HAMI","url":"https://www.nature.com/articles/s41598-025-10586-x","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Hippocampus-inspired RL framework","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"graph-traversal|similarity","persistence":"long-term","update":"consolidation","unit":"episode","governance":"opaque","conflict":"llm-arbitrate","description":"Symbolic indexing + hierarchical memory refinement + structured episodic retrieval, inspired by hippocampal mechanisms. Nature Scientific Reports 2025.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/hami--nature-com"},{"id":"hanabi-learning-environment--gh-google-deepmind-hanabi-learning-environment","name":"Hanabi Learning Environment","url":"https://github.com/google-deepmind/hanabi-learning-environment","section":"Memory benchmarks & evaluation","tier":3,"status":"archived","last_verified_at":"2026-10-28","type":"Cooperative multi-agent partial-observability card-game benchmark","license":"Apache-2.0","latest_release":"","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"700★ C++/Python","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"DeepMind's Hanabi Learning Environment — a benchmark for cooperative multi-agent agents under partial observability and theory-of-mind. Players see each other's cards but not their own; success requires modeling teammates' beliefs. Bard (Bowling et al. 2020) described Hanabi as 'a new frontier for AI research'. Used in OpenAI ad hoc team-play papers and DeepMind's Bayesian Action Decoder.","cited_cells":29,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/hanabi-learning-environment--gh-google-deepmind-hanabi-learning-environment"},{"id":"hela-mem--arxiv-2604-16839","name":"HeLa-Mem","url":"https://arxiv.org/abs/2604.16839","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Hebbian distillation + hippocampal-cortical consolidation","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"graph","retrieval":"graph-traversal","persistence":"long-term","update":"consolidation","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Co-activation strengthens associative edges (Hebbian Distillation); separate consolidation pass periodically extracts semantic summaries from the episodic graph — directly modelling slow-wave hippocampal-to-cortical consolidation. Dual-path retrieval over both stores.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/hela-mem--arxiv-2604-16839"},{"id":"helmet--arxiv-2410-02694","name":"HELMET","url":"https://arxiv.org/abs/2410.02694","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Holistic downstream-task long-context eval","license":"MIT","latest_release":"no-release no-release","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"214★ +2/mo Jupyter Notebook","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Princeton NLP. 7 task categories — RAG with real retrieval, cited generation, summarisation, re-ranking, classification, many-shot ICL, long-document QA — at lengths up to 128K. Model-based evaluation enables open-ended scoring.","cited_cells":30,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/helmet--arxiv-2410-02694"},{"id":"human-like-remembering-and-forgetting--dl-acm-org","name":"Human-Like Remembering and Forgetting","url":"https://dl.acm.org/doi/10.1145/3765766.3765803","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"ACT-R Base-Level Activation over vector store","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"consolidation","unit":"chunk","governance":"inspectable","conflict":"llm-arbitrate","description":"Re-implements ACT-R's retrieval probability model — temporal decay, prior-access frequency, stochastic noise — as scoring on top of a vector store. Spreading activation over semantic similarity adds associative recall. Forgetting becomes an explicit function of recency and use-count, not eviction heuristics.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/human-like-remembering-and-forgetting--dl-acm-org"},{"id":"hyena--arxiv-2302-10866","name":"Hyena","url":"https://arxiv.org/abs/2302.10866","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Long-convolution alternative to attention","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"kv","retrieval":"similarity","persistence":"session","update":"replacement","unit":"token","governance":"opaque","conflict":"n/a","description":"Subquadratic alternative to attention — implicit long convolutions + data-controlled gating. Stanford. Ancestor of StripedHyena.","cited_cells":6,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/hyena--arxiv-2302-10866"},{"id":"i-jepa--arxiv-2301-08243","name":"I-JEPA","url":"https://arxiv.org/abs/2301.08243","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"archived","last_verified_at":"2026-10-23","type":"Joint-embedding predictive arch (image)","license":"CC-BY-NC-4.0","latest_release":"no releases","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"3.4k★ Python","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"read-only","unit":"scene-graph","governance":"opaque","conflict":"n/a","description":"Meta. Self-supervised image learning by predicting target-block representations from a context block in latent space — no hand-crafted augmentations or pixel reconstruction. Foundation of LeCun's hierarchical-JEPA programme. CVPR 2023.","cited_cells":23,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/i-jepa--arxiv-2301-08243"},{"id":"ibm-meta-tokens--arxiv-2506-00307","name":"IBM Meta-Tokens","url":"https://arxiv.org/abs/2506.00307","section":"Recent method papers — theorized, no distinct product","tier":2,"status":"active","last_verified_at":"2026-10-28","type":"Lossless context compression via learned placeholders","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"searched not found","llm_lock":"","github":"","storage":"parametric|kv-cache","retrieval":"attention|cache-lookup","persistence":"parametric-permanent","update":"parametric-edit","unit":"kv-token|weight","governance":"opaque","conflict":"n/a","description":"Meta-tokens are learned vocabulary entries that act as placeholders for recurring token subsequences — the model is fine-tuned to understand compressed input where meta-tokens stand in for their corresponding subsequences without any information loss. Mathematically lossless and reversible, unlike soft-compression approaches.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/ibm-meta-tokens--arxiv-2506-00307"},{"id":"iclr-2026-memagents-workshop--sites-google-com","name":"ICLR 2026 MemAgents Workshop","url":"https://sites.google.com/view/memagent-iclr26/","section":"Memory benchmarks & evaluation","tier":2,"status":"active","last_verified_at":"2026-10-23","type":"Workshop venue — RL + cognitive psychology + LLM memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"none-trivial","retrieval":"none","persistence":"n/a","update":"n/a","unit":"n/a","governance":"n/a","conflict":"out-of-scope","description":"Full-day hybrid workshop held April 27, 2026, Rio de Janeiro. Scope: episodic and semantic memory, working memory, parametric knowledge, KGs, vector DBs, retrieval pipelines, context management, long-context utilization, temporal credit assignment. Bridges RL, neuroscience, and cognitive psychology alongside the engineering track.","cited_cells":14,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/iclr-2026-memagents-workshop--sites-google-com"},{"id":"implicitmembench--arxiv-2604-08064","name":"ImplicitMemBench","url":"https://arxiv.org/abs/2604.08064","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Implicit / procedural memory benchmark","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"300 items across three paradigms: procedural memory (one-shot skill retention after interference), priming (theme-driven response biases), classical conditioning (CS-US exposure shaping first decisions). Unified learning-interference-test protocol with first-attempt scoring to isolate automatized behavior — distinct from explicit-recall benchmarks.","cited_cells":16,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/implicitmembench--arxiv-2604-08064"},{"id":"inducing-programmatic-skills--arxiv-2504-06821","name":"Inducing Programmatic Skills","url":"https://arxiv.org/abs/2504.06821","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Programmatic skill induction","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"file","retrieval":"similarity","persistence":"long-term","update":"extraction","unit":"skill","governance":"inspectable","conflict":"llm-arbitrate","description":"Inducing programmatic skills for agentic task enhancement.","cited_cells":22,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/inducing-programmatic-skills--arxiv-2504-06821"},{"id":"infini-attention--arxiv-2404-07143","name":"Infini-attention","url":"https://arxiv.org/abs/2404.07143","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"research-code","last_verified_at":"2026-10-28","type":"Compressive memory + linear attention","license":"Unlicensed","latest_release":"no releases","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"55★ Python","storage":"kv-cache","retrieval":"attention","persistence":"session","update":"consolidation","unit":"kv-token","governance":"opaque","conflict":"n/a","description":"Google. \"Leave No Context Behind.\" Compressive memory inside the attention block + masked local + long-term linear attention.","cited_cells":24,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/infini-attention--arxiv-2404-07143"},{"id":"infinitebench-bench--arxiv-2402-13718","name":"InfiniteBench (∞Bench)","url":"https://arxiv.org/abs/2402.13718","section":"Memory benchmarks & evaluation","tier":3,"status":"research-code","last_verified_at":"2026-10-28","type":"100K+ token realistic long-context comprehension","license":"MIT","latest_release":"no releases","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"382★ Python","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Tsinghua / OpenBMB. First benchmark with average input >100K tokens. 12 tasks across English + Chinese: long novel QA, code debugging, math in long context, summarisation, retrieval. Realistic documents (novels, code) rather than synthetic filler.","cited_cells":29,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/infinitebench-bench--arxiv-2402-13718"},{"id":"iterresearch--arxiv-2511-07327","name":"IterResearch","url":"https://arxiv.org/abs/2511.07327","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Markovian state reconstruction","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv","retrieval":"agentic","persistence":"session","update":"overwrite","unit":"summary","governance":"inspectable","conflict":"llm-arbitrate","description":"Rethinks long-horizon agents via Markovian state reconstruction.","cited_cells":20,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/iterresearch--arxiv-2511-07327"},{"id":"jarvis-1","name":"JARVIS-1","url":"","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Open-world memory-augmented agents","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"cross-session","update":"extraction","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Open-world multi-task agents with memory-augmented multimodal LLMs. TPAMI 2024 . (Distinct from any other JARVIS.)","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/jarvis-1"},{"id":"jepa-six-module-world-model-informal--openreview-net","name":"JEPA Six-Module World Model (informal)","url":"https://openreview.net/pdf?id=BZ5a1r-kVsf","section":"Theoretical / informal — ideas without a paper","tier":5,"status":"active","last_verified_at":"2026-05-14","type":"Memory as architecturally peer module","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"weight","governance":"opaque","conflict":"n/a","description":"Six-module cognitive architecture (perception, world model, cost, memory, action, configurator). Memory is a distinct addressable module with its own update rules, separate from the world model that reads/updates it — not in-context storage, not RAG. Directly challenges the \"memory = RAG\" consensus.","cited_cells":11,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/jepa-six-module-world-model-informal--openreview-net"},{"id":"knowledge-graph-tuning--arxiv-2405-19686","name":"Knowledge Graph Tuning","url":"https://arxiv.org/abs/2405.19686","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Real-time KG personalisation","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"graph","retrieval":"graph-traversal","persistence":"long-term","update":"overwrite","unit":"fact","governance":"user-controllable","conflict":"llm-arbitrate","description":"Real-time LLM personalisation based on human feedback.","cited_cells":21,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/knowledge-graph-tuning--arxiv-2405-19686"},{"id":"larimar--arxiv-2403-11901","name":"Larimar","url":"https://arxiv.org/abs/2403.11901","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Distributed hippocampal episodic memory for editing","license":"Apache-2.0","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"36★ Python","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"episode","governance":"editable","conflict":"overwrite","description":"IBM Research. Attaches a fast-learning distributed episodic memory (hippocampus-inspired) to a frozen LLM, enabling one-shot knowledge updates without retraining. Supports selective forgetting + information-leakage prevention.","cited_cells":26,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/larimar--arxiv-2403-11901"},{"id":"learnact--arxiv-2504-13805","name":"LearnAct","url":"https://arxiv.org/abs/2504.13805","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Few-shot mobile GUI agent","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv","retrieval":"injection","persistence":"session","update":"read-only","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Few-shot mobile GUI agent with unified demonstration benchmark.","cited_cells":19,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/learnact--arxiv-2504-13805"},{"id":"legomem--arxiv-2510-04851","name":"LEGOMem","url":"https://arxiv.org/abs/2510.04851","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Modular procedural memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity","persistence":"long-term","update":"extraction","unit":"skill","governance":"inspectable","conflict":"llm-arbitrate","description":"Modular procedural memory for multi-agent LLM systems.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/legomem--arxiv-2510-04851"},{"id":"licomemory--arxiv-2511-01448","name":"LiCoMemory","url":"https://arxiv.org/abs/2511.01448","section":"Research / specialised systems","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"CogniGraph + temporal/hierarchy-aware retrieval","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"graph","retrieval":"graph-traversal|similarity","persistence":"long-term","update":"extraction","unit":"episode","governance":"opaque","conflict":"llm-arbitrate","description":"Lightweight hierarchical graph (CogniGraph) with entities and relations as semantic indexing layers. Incremental graph construction, fast updates, low-latency inference. Nov 2025.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/licomemory--arxiv-2511-01448"},{"id":"lightmem--arxiv-2510-18866","name":"LightMem","url":"https://arxiv.org/abs/2510.18866","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Lightweight memory-augmented","license":"MIT","latest_release":"no releases","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"820★ +30/mo Python","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"extraction","unit":"fact","governance":"inspectable","conflict":"none","description":"Lightweight, efficient memory-augmented generation.","cited_cells":24,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/lightmem--arxiv-2510-18866"},{"id":"llm-actr-cognitive-llms--arxiv-2408-09176","name":"LLM-ACTR / Cognitive LLMs","url":"https://arxiv.org/abs/2408.09176","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"ACT-R latent representations injected into LLM","license":"Unlicensed","latest_release":"no releases","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"5★ Jupyter Notebook","storage":"parametric","retrieval":"parametric-recall|similarity","persistence":"parametric-permanent","update":"read-only","unit":"chunk","governance":"opaque","conflict":"n/a","description":"Neuro-symbolic architecture for manufacturing decision-making. Extracts ACT-R's procedural / declarative / goal buffers as latent neural representations, then injects them into trainable LLM adapter layers via knowledge transfer.","cited_cells":21,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/llm-actr-cognitive-llms--arxiv-2408-09176"},{"id":"llm-as-os-kernel--x-com","name":"LLM as OS Kernel","url":"https://x.com/karpathy/status/1707437820045062561","section":"Theoretical / informal — ideas without a paper","tier":5,"status":"active","last_verified_at":"2026-10-28","type":"Architectural metaphor","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"n/a","unit":"file","governance":"n/a","conflict":"n/a","description":"LLM is \"the kernel process of a new Operating System\" orchestrating I/O, code interpreter, browser, and an embedding store for files / internal memory. Context window = RAM; embedding store = filesystem.","cited_cells":12,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/llm-as-os-kernel--x-com"},{"id":"llm-os-specs--x-com","name":"LLM OS Specs","url":"https://x.com/karpathy/status/1723140519554105733","section":"Theoretical / informal — ideas without a paper","tier":5,"status":"active","last_verified_at":"2026-10-28","type":"Memory in hardware terms","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"n/a","unit":"file","governance":"n/a","conflict":"n/a","description":"Operationalises the kernel analogy: 256-core processor at 20 Hz (tokens/sec), RAM = 128K tokens of context, filesystem = Ada-002 embedding store. Memory capacity becomes a measurable architectural constraint.","cited_cells":12,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/llm-os-specs--x-com"},{"id":"llm-wiki-compiled-memory--gh-karpathy-442a6bf555914893e9891c11519de94f","name":"LLM Wiki — compiled memory","url":"https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f","section":"Theoretical / informal — ideas without a paper","tier":5,"status":"research-code","last_verified_at":"2026-10-28","type":"Compilation over RAG","license":"not applicable — informal artifact (no formal license)","latest_release":"not applicable — informal artifact (gist/tweet/blog)","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"0★ no data — repo unreachable","storage":"file","retrieval":"injection","persistence":"long-term","update":"overwrite","unit":"file","governance":"editable","conflict":"n/a","description":"Three-layer directory: raw/ (immutable sources), wiki/ (LLM-maintained compiled pages), CLAUDE.md (schema). LLM reads sources and integrates them into the wiki — updating pages, resolving contradictions — so knowledge compounds across sessions instead of being retrieved cold.","cited_cells":14,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/llm-wiki-compiled-memory--gh-karpathy-442a6bf555914893e9891c11519de94f"},{"id":"lm2--arxiv-2502-06049","name":"LM2","url":"https://arxiv.org/abs/2502.06049","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Large memory models","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"attention|similarity","persistence":"long-term","update":"parametric-edit","unit":"weight","governance":"opaque","conflict":"n/a","description":"Large memory models — explicit external memory module within transformer.","cited_cells":20,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/lm2--arxiv-2502-06049"},{"id":"locomo--arxiv-2402-17753","name":"LoCoMo","url":"https://arxiv.org/abs/2402.17753","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Very long-term multi-session dialogue memory","license":"MIT (paper code release)","latest_release":"no-release no-release","deployment":"not applicable — evaluation harness; run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"39★ Jupyter Notebook","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Conversations spanning up to 35 sessions / 300 turns / ~9K tokens, built from LLM-agent personas grounded on temporal event graphs and verified by human annotators. Tests QA (single-hop, multi-hop, temporal, commonsense, adversarial), event-graph summarisation, multimodal dialogue.","cited_cells":29,"third_party_cells":6,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/locomo--arxiv-2402-17753"},{"id":"longbench--arxiv-2308-14508","name":"LongBench","url":"https://arxiv.org/abs/2308.14508","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Bilingual multitask long-context suite","license":"MIT","latest_release":"no releases","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"1.2k★ Python","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Tsinghua THUDM. 21 datasets across 6 categories (single-doc QA, multi-doc QA, summarisation, few-shot ICL, synthetic, code completion) in English + Chinese. ~6.7K words English / 13.4K chars Chinese average. v2 at ACL 2025 with harder, longer instances.","cited_cells":29,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/longbench--arxiv-2308-14508"},{"id":"longlora--arxiv-2309-12307","name":"LongLoRA","url":"https://arxiv.org/abs/2309.12307","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Sparse-attention LoRA for long context","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"kv-cache","retrieval":"attention","persistence":"session","update":"append-only|parametric-edit","unit":"kv-token|weight","governance":"opaque","conflict":"out-of-scope","description":"Shift-short attention + LoRA fine-tuning that extends LLaMA context length cheaply — 100k tokens on a single GPU. ICLR 2024.","cited_cells":6,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/longlora--arxiv-2309-12307"},{"id":"longmem--arxiv-2306-07174","name":"LongMem","url":"https://arxiv.org/abs/2306.07174","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"research-code","last_verified_at":"2026-10-28","type":"Trainable SideNet over memory bank","license":"Apache-2.0","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"826★ Python","storage":"kv-cache","retrieval":"similarity|attention","persistence":"long-term","update":"append-only","unit":"kv-token","governance":"opaque","conflict":"n/a","description":"Trainable SideNet module selectively retrieves relevant cached key-value pairs from memory banks during decoding.","cited_cells":27,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/longmem--arxiv-2306-07174"},{"id":"longmemeval--arxiv-2410-10813","name":"LongMemEval","url":"https://arxiv.org/abs/2410.10813","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Long-term chat-assistant memory across sessions","license":"MIT","latest_release":"no releases","deployment":"not applicable — evaluation harness; run locally against model API","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"740★ Python","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"5 core memory abilities — info extraction, multi-session reasoning, temporal reasoning, knowledge updates, abstention — via 500 curated questions in synthetic chat histories. Two scales: LongMemEval-S (115K tokens), LongMemEval-M (up to 1.5M).","cited_cells":28,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/longmemeval--arxiv-2410-10813"},{"id":"lyfe-agents--arxiv-2310-02172","name":"Lyfe Agents","url":"https://arxiv.org/abs/2310.02172","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Generative social agents","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"session","update":"extraction","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Generative agents for low-cost real-time social interactions.","cited_cells":20,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/lyfe-agents--arxiv-2310-02172"},{"id":"m-memoryllm--arxiv-2502-00592","name":"M+ (MemoryLLM)","url":"https://arxiv.org/abs/2502.00592","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Scalable long-term latent memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"weight","governance":"opaque","conflict":"n/a","description":"Extends MemoryLLM with scalable long-term memory.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/m-memoryllm--arxiv-2502-00592"},{"id":"magma--arxiv-2601-03236","name":"MAGMA","url":"https://arxiv.org/abs/2601.03236","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Multi-graph agentic memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"graph","retrieval":"graph-traversal|similarity","persistence":"long-term","update":"extraction","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Multi-graph agentic memory architecture for AI agents.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/magma--arxiv-2601-03236"},{"id":"mamba--arxiv-2312-00752","name":"Mamba","url":"https://arxiv.org/abs/2312.00752","section":"Recent method papers — theorized, no distinct product","tier":2,"status":"active","last_verified_at":"2026-10-28","type":"Selective state-space model","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"kv","retrieval":"similarity","persistence":"session","update":"replacement","unit":"token","governance":"opaque","conflict":"n/a","description":"Selective state-space model — input-dependent SSM with linear scaling; competitive with transformers at smaller scales. CMU.","cited_cells":5,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/mamba--arxiv-2312-00752"},{"id":"mamba-2-ssd--arxiv-2405-21060","name":"Mamba-2 / SSD","url":"https://arxiv.org/abs/2405.21060","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Structured State-space Duality","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"kv","retrieval":"attention","persistence":"session","update":"replacement","unit":"token","governance":"opaque","conflict":"out-of-scope","description":"Structured State-space Duality — Mamba-2 architecture that unifies SSMs and attention; 2-8× faster than Mamba-1. ICML 2024.","cited_cells":6,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/mamba-2-ssd--arxiv-2405-21060"},{"id":"matter--arxiv-2406-04670","name":"MATTER","url":"https://arxiv.org/abs/2406.04670","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Heterogeneous-source neural memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity|attention","persistence":"long-term","update":"read-only","unit":"document","governance":"opaque","conflict":"n/a","description":"Memory-Augmented Transformer Using Heterogeneous Knowledge Sources. Unifies unstructured paragraphs + semi-structured QA pairs into fixed-length type-agnostic neural memories. Jun 2024.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/matter--arxiv-2406-04670"},{"id":"mem--arxiv-2509-25911","name":"Mem-α","url":"https://arxiv.org/abs/2509.25911","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"RL-learned memory construction","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"agent-controlled","unit":"episode","governance":"agent-controlled","conflict":"llm-arbitrate","description":"Learning memory construction via reinforcement learning.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/mem--arxiv-2509-25911"},{"id":"mem1--arxiv-2506-15841","name":"MEM1","url":"https://arxiv.org/abs/2506.15841","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Synergised memory + reasoning","license":"MIT","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"310★ +2/mo Python","storage":"kv-cache","retrieval":"attention","persistence":"session","update":"consolidation","unit":"summary","governance":"opaque","conflict":"llm-arbitrate","description":"Learns to synergise memory and reasoning for efficient agents.","cited_cells":27,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/mem1--arxiv-2506-15841"},{"id":"memagent--arxiv-2507-02259","name":"MemAgent","url":"https://arxiv.org/abs/2507.02259","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"RL multi-conv memory agent","license":"Apache-2.0","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"1.0k★ +10/mo Python","storage":"kv","retrieval":"agentic","persistence":"long-term","update":"agent-controlled","unit":"summary","governance":"opaque","conflict":"llm-arbitrate","description":"Reshapes long-context LLM with multi-conv RL-based memory agent.","cited_cells":25,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memagent--arxiv-2507-02259"},{"id":"membart","name":"MemBART","url":"","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Memory states across dialogue turns","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv-cache","retrieval":"attention","persistence":"long-term","update":"overwrite","unit":"kv-token","governance":"opaque","conflict":"n/a","description":"Preserves memory states across dialogue turns for coherent long-term interactions. (Cited in survey but the paper's exact arxiv/venue identifier was not recoverable in a quick search — likely a less-prominent dialogue-systems paper.)","cited_cells":13,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/membart"},{"id":"membench--arxiv-2506-21605","name":"MemBench","url":"https://arxiv.org/abs/2506.21605","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Multi-dim agent memory mechanism comparison","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Renmin University. Evaluates LLM-agent memory mechanisms across three axes — effectiveness (accuracy), efficiency (LLM call count), capacity (degradation as memory grows). Distinguishes factual vs reflective memory, participation vs observation scenarios.","cited_cells":15,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/membench--arxiv-2506-21605"},{"id":"memconflict--arxiv-2605-20926","name":"MemConflict","url":"https://arxiv.org/abs/2605.20926","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Benchmark","license":"unspecified","latest_release":"last commit 2026-06-27","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/TaoZhen1110/MemConflict (13★)","storage":"none-trivial","retrieval":"none","persistence":"cross-session","update":"read-only","unit":"turn|fact","governance":"n/a","conflict":"out-of-scope","description":"Diagnostic benchmark covering dynamic, static and conditional memory conflicts, with black-box answer and white-box retrieval/ranking metrics; it evaluates A-Mem, LangMem, Letta, MemOS, Mem0 and Memobase. Traction: Head-to-head evaluation of six named memory systems; Renmin Univ + MemTensor","cited_cells":10,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memconflict--arxiv-2605-20926"},{"id":"memento--arxiv-2508-16153","name":"Memento","url":"https://arxiv.org/abs/2508.16153","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Behavior tuning without weight updates","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"extraction","unit":"episode","governance":"agent-controlled","conflict":"llm-arbitrate","description":"Fine-tunes LLM-agent behavior without fine-tuning the underlying model.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memento--arxiv-2508-16153"},{"id":"memevobench--arxiv-2604-15774","name":"MemEvoBench","url":"https://arxiv.org/abs/2604.15774","section":"Memory benchmarks & evaluation","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Adversarial memory-evolution safety benchmark","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"QA tasks across 7 domains / 36 risk types plus workflow tasks adapted from 20 Agent-SafetyBench environments with noisy tool returns; both use mixed benign-and-misleading memory pools in multi-round interactions. Targets memory-evolution safety, not recall accuracy — first benchmark specifically targeting adversarial degradation of evolving memory.","cited_cells":15,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memevobench--arxiv-2604-15774"},{"id":"memevolve--arxiv-2512-18746","name":"MemEvolve","url":"https://arxiv.org/abs/2512.18746","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Meta-evolution of memory systems","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"file","retrieval":"agentic","persistence":"cross-session","update":"consolidation","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Meta-evolution of agent memory systems. Includes EvolveLab — a unified self-evolving memory codebase distilling 12 representative memory systems.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memevolve--arxiv-2512-18746"},{"id":"memformer","name":"Memformer","url":"","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"research-code","last_verified_at":"2026-10-28","type":"External KV + Memory Replay Backprop","license":"MIT","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"9★ Python","storage":"kv-cache","retrieval":"attention|similarity","persistence":"session","update":"consolidation","unit":"kv-token","governance":"opaque","conflict":"n/a","description":"External key-value stores with similarity-based cache management; trained with Memory Replay Backpropagation.","cited_cells":26,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memformer"},{"id":"memformers-gradient-memory","name":"Memformers (gradient memory)","url":"","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Past-gradient memory registers","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"attention","persistence":"session","update":"parametric-edit","unit":"weight","governance":"opaque","conflict":"n/a","description":"Treats past optimisation gradients as memory registers for procedural computation. Distinct from the 2010.06891 \"Memformer\" above. (Survey citation — distinct paper identifier not recovered.)","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memformers-gradient-memory"},{"id":"memgen--arxiv-2509-24704","name":"MemGen","url":"https://arxiv.org/abs/2509.24704","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Generative latent memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"weight","governance":"opaque","conflict":"llm-arbitrate","description":"Weaves generative latent memory for self-evolving agents.","cited_cells":20,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memgen--arxiv-2509-24704"},{"id":"memgpt-v2-agent-tools--arxiv-2310-08560","name":"MemGPT v2 / agent-tools","url":"https://arxiv.org/abs/2310.08560","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"OS-style hierarchical memory","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"hybrid|vector|kv","retrieval":"agentic|similarity|injection","persistence":"long-term","update":"agent-controlled","unit":"episode|summary|fact","governance":"inspectable","conflict":"agent","description":"Already in catalog as the foundational MemGPT paper. Note: Letta is the productionised successor (cross-listed).","cited_cells":5,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memgpt-v2-agent-tools--arxiv-2310-08560"},{"id":"memguide--arxiv-2505-20231","name":"MemGuide","url":"https://arxiv.org/abs/2505.20231","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Intent-driven memory selection","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"cross-session","update":"extraction","unit":"fact","governance":"inspectable","conflict":"llm-arbitrate","description":"Intent-driven memory selection for goal-oriented multi-session agents.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memguide--arxiv-2505-20231"},{"id":"meminsight--arxiv-2503-21760","name":"MemInsight","url":"https://arxiv.org/abs/2503.21760","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Autonomous semantic augmentation","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"extraction","unit":"summary","governance":"n/a","conflict":"llm-arbitrate","description":"Autonomously augments semantic representations of historical interactions to improve retrieval quality without manual curation. Validated on conversational recommendation, QA, and event summarisation. EMNLP 2025 Main.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/meminsight--arxiv-2503-21760"},{"id":"memlong--arxiv-2408-16039","name":"MemLong","url":"https://arxiv.org/abs/2408.16039","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Retrieval-causal attention + pruning","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv-cache","retrieval":"attention|similarity","persistence":"long-term","update":"evict-oldest","unit":"kv-token","governance":"n/a","conflict":"n/a","description":"Retrieval Causal Attention actively prunes less important cached entries.","cited_cells":21,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memlong--arxiv-2408-16039"},{"id":"memlora--arxiv-2512-04763","name":"MemLoRA","url":"https://arxiv.org/abs/2512.04763","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Distilled expert adapters","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"skill","governance":"n/a","conflict":"n/a","description":"Distils expert adapters for on-device memory systems.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memlora--arxiv-2512-04763"},{"id":"memmachine-paper--arxiv-2604-04853","name":"MemMachine (paper)","url":"https://arxiv.org/abs/2604.04853","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Ground-truth-preserving episodic + cluster-expansion retrieval","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"append-only","unit":"episode","governance":"inspectable","conflict":"append","description":"Stores raw conversational episodes without LLM-driven extraction, indexing at sentence level; retrieval expands nucleus matches into neighboring-context clusters to address the embedding-dissimilarity problem in dialogue data. A Retrieval Agent routes queries across direct, parallel-decomposition, or iterative chain-of-query strategies. Distinct from the MemMachine framework already listed above (Dedicated memory layers).","cited_cells":20,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memmachine-paper--arxiv-2604-04853"},{"id":"memobrain--arxiv-2601-08079","name":"MemoBrain","url":"https://arxiv.org/abs/2601.08079","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Dependency-aware working-context compaction","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv","retrieval":"agentic","persistence":"session","update":"consolidation","unit":"chunk","governance":"n/a","conflict":"llm-arbitrate","description":"Co-pilot alongside reasoning agent. Builds dependency graph over reasoning steps; prunes invalid steps, folds completed sub-trajectories, preserves compact high-salience backbone under fixed context budget — without blocking execution.","cited_cells":20,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memobrain--arxiv-2601-08079"},{"id":"memochat--arxiv-2308-08239","name":"MemoChat","url":"https://arxiv.org/abs/2308.08239","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"research-code","last_verified_at":"2026-10-28","type":"Memos for long-range conversation","license":"MIT","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"28★ Python","storage":"kv","retrieval":"similarity","persistence":"long-term","update":"extraction","unit":"summary","governance":"n/a","conflict":"llm-arbitrate","description":"Tunes LLMs to use memos for consistent long-range conversation.","cited_cells":24,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memochat--arxiv-2308-08239"},{"id":"memolet--doi-org","name":"Memolet","url":"https://doi.org/10.1145/3654777.3676388","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"User-AI conversational memory reuse","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"cross-session","update":"append-only","unit":"chunk","governance":"user-controllable","conflict":"manual","description":"Reifies the reuse of user-AI conversational memories. CHI/UIST.","cited_cells":19,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memolet--doi-org"},{"id":"memorag--gh-qhjqhj00-memorag","name":"MemoRAG","url":"https://github.com/qhjqhj00/MemoRAG","section":"Research / specialised systems","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Global memory-enhanced RAG","license":"Apache-2.0","latest_release":"v0.1.5 2024-09-24","deployment":"searched not found","pricing":"searched not found","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"2.2k★ Python","storage":"vector","retrieval":"similarity|hybrid-rerank","persistence":"long-term","update":"read-only","unit":"chunk","governance":"inspectable","conflict":"n/a","description":"RAG framework on top of a long-context memory model. Builds global memory once, generates contextual clues at query time. TheWebConf 2025.","cited_cells":26,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memorag--gh-qhjqhj00-memorag"},{"id":"memoria-cognitive--arxiv-2310-03052","name":"Memoria (cognitive)","url":"https://arxiv.org/abs/2310.03052","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Human-inspired memory architecture","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity","persistence":"long-term","update":"consolidation","unit":"episode","governance":"n/a","conflict":"llm-arbitrate","description":"Resolves fateful forgetting through human-inspired memory architecture. Distinct from MatrixOrigin's commercial Memoria.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memoria-cognitive--arxiv-2310-03052"},{"id":"memorizing-transformers--openreview-net","name":"Memorizing Transformers","url":"https://openreview.net/forum?id=TrjbxzRcnf-","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-05-14","type":"External memory in transformers","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv-cache","retrieval":"similarity|attention","persistence":"session","update":"append-only","unit":"kv-token","governance":"n/a","conflict":"n/a","description":"Architecture with external retrieval memory in the transformer.","cited_cells":22,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memorizing-transformers--openreview-net"},{"id":"memoro--doi-org","name":"Memoro","url":"https://doi.org/10.1145/3613904.3642450","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Real-time memory augmentation UI","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"session","update":"read-only","unit":"chunk","governance":"inspectable","conflict":"n/a","description":"Concise interface for real-time memory augmentation. CHI.","cited_cells":17,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memoro--doi-org"},{"id":"memory--arxiv-2407-01178","name":"Memory³","url":"https://arxiv.org/abs/2407.01178","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Explicit memory in LLM","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall|similarity","persistence":"parametric-permanent","update":"parametric-edit","unit":"document","governance":"inspectable","conflict":"n/a","description":"Language modeling with explicit memory. Landmark.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memory--arxiv-2407-01178"},{"id":"memory-as-action--arxiv-2510-12635","name":"Memory as Action","url":"https://arxiv.org/abs/2510.12635","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Autonomous context curation","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"agentic","persistence":"long-term","update":"agent-controlled","unit":"episode","governance":"agent-controlled","conflict":"llm-arbitrate","description":"Autonomous context curation for long-horizon agentic tasks.","cited_cells":16,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memory-as-action--arxiv-2510-12635"},{"id":"memory-as-metabolism--arxiv-2604-12034","name":"Memory as Metabolism","url":"https://arxiv.org/abs/2604.12034","section":"Theoretical / informal — ideas without a paper","tier":5,"status":"active","last_verified_at":"2026-10-23","type":"Metabolic governance cycle for companion knowledge","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"n/a","retrieval":"n/a","persistence":"lifelong","update":"consolidation|evict-oldest","unit":"document","governance":"auditable","conflict":"append","description":"Argues that companion AI knowledge systems must treat memory as a continuous metabolic process rather than a static store. Proposes a three-tier store (raw buffer, active wiki, cold memory) governed by five operations (TRIAGE, DECAY, CONTEXTUALIZE, CONSOLIDATE, AUDIT), memory gravity (recency/frequency weighting), and explicit minority-hypothesis retention to counter worldview entrenchment. April 13, 2026.","cited_cells":12,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memory-as-metabolism--arxiv-2604-12034"},{"id":"memory-decoder--arxiv-2508-09874","name":"Memory Decoder","url":"https://arxiv.org/abs/2508.09874","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Pretrained plug-and-play memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"read-only","unit":"document","governance":"n/a","conflict":"n/a","description":"Pretrained, plug-and-play memory for LLMs.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memory-decoder--arxiv-2508-09874"},{"id":"memory-layers-at-scale--arxiv-2412-09764","name":"Memory Layers at Scale","url":"https://arxiv.org/abs/2412.09764","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Trainable KV layers + product-key","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"similarity","persistence":"parametric-permanent","update":"parametric-edit","unit":"weight","governance":"n/a","conflict":"n/a","description":"Meta. Replaces dense feed-forward blocks with trainable key-value layers using product-key lookup.","cited_cells":17,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memory-layers-at-scale--arxiv-2412-09764"},{"id":"memory-poisoning-as-attack-surface--simonwillison-net","name":"Memory Poisoning as Attack Surface","url":"https://simonwillison.net/2025/Sep/18/agents/","section":"Theoretical / informal — ideas without a paper","tier":5,"status":"active","last_verified_at":"2026-10-23","type":"Security framing","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"n/a","retrieval":"agentic","persistence":"long-term","update":"agent-controlled","unit":"fact|attack","governance":"n/a","conflict":"out-of-scope","description":"Agent persistent memory is an attack surface: an adversarial document in the environment can instruct the agent's memory tool to write malicious facts, poisoning the long-term store. The more capable the memory system, the larger the injection surface. Conversation history is memory's first-class form; long-term memory is \"an extra set of tools.\"","cited_cells":11,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memory-poisoning-as-attack-surface--simonwillison-net"},{"id":"memory-r1--aclanthology-org","name":"Memory-R1","url":"https://aclanthology.org/2026.acl-long.583/","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Paper","license":"unknown","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"","storage":"vector","retrieval":"similarity|agentic","persistence":"long-term","update":"agent-controlled|extraction","unit":"fact","governance":"plaintext","conflict":"llm-arbitrate","description":"Uses RL (PPO/GRPO) to train a Memory Manager that learns ADD/UPDATE/DELETE/NOOP operations and an Answer Agent that filters retrieved memories. Beats heuristic memory baselines on LoCoMo, MSC and LongMemEval from just 152 training QA pairs. Traction: ACL 2026 long paper; LMU/Oxford/TUM; widely cited RL-for-memory work","cited_cells":6,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memory-r1--aclanthology-org"},{"id":"memory3--arxiv-2407-01178","name":"Memory3","url":"https://arxiv.org/abs/2407.01178","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Explicit memory in LLM third memory","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"kv","retrieval":"similarity","persistence":"long-term","update":"accretion","unit":"token","governance":"opaque","conflict":"n/a","description":"Three-tier memory model: implicit (weights), working (context), and explicit (compressed KV stored externally). 2.4B model competitive with 7B+ using explicit memory.","cited_cells":5,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memory3--arxiv-2407-01178"},{"id":"memoryagentbench--arxiv-2507-05257","name":"MemoryAgentBench","url":"https://arxiv.org/abs/2507.05257","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"4-competency agent memory benchmark","license":"Unlicensed","latest_release":"no releases","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use (unlicensed GitHub)","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"323★ Python","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"HUST. Reconstructs existing datasets into incremental multi-turn dialogue chunks fed in time order. Plus EventQA (accurate retrieval) and FactConsolidation (selective forgetting). Targets retrieval, test-time learning, long-range understanding, selective forgetting.","cited_cells":26,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memoryagentbench--arxiv-2507-05257"},{"id":"memoryarena--arxiv-2602-16313","name":"MemoryArena","url":"https://arxiv.org/abs/2602.16313","section":"Memory benchmarks & evaluation","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Multi-session agentic memory eval gym","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Stanford Digital Economy Lab et al. Benchmarks memory in realistic Memory-Agent-Environment loops where tasks span multiple sessions and subtasks are explicitly interdependent — agents must distill prior-session experiences into memory and use that memory to act later. Exposes the gap between conversational memory benchmarks (LoCoMo) and true agentic memory.","cited_cells":15,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memoryarena--arxiv-2602-16313"},{"id":"memorybank--arxiv-2305-10250","name":"MemoryBank","url":"https://arxiv.org/abs/2305.10250","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"research-code","last_verified_at":"2026-10-23","type":"Long-term memory enhancement","license":"MIT","latest_release":"LoRA_checkpoint 2023-05-21","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"427★ Python","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"consolidation","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Enhances LLMs with long-term memory. Early influential paper.","cited_cells":25,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memorybank--arxiv-2305-10250"},{"id":"memorybench-tsinghua-thuir--arxiv-2510-17281","name":"MemoryBench (Tsinghua THUIR)","url":"https://arxiv.org/abs/2510.17281","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Declarative + procedural continual-learning bench","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Tsinghua THUIR. First benchmark covering both declarative and procedural memory for LLM systems. User-feedback simulation framework tests adaptive improvement. Distinct from Renmin's MemBench.","cited_cells":15,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memorybench-tsinghua-thuir--arxiv-2510-17281"},{"id":"memoryllm-original--arxiv-2402-04624","name":"MemoryLLM (original)","url":"https://arxiv.org/abs/2402.04624","section":"Recent method papers — theorized, no distinct product","tier":2,"status":"active","last_verified_at":"2026-10-28","type":"Fixed-size in-weights memory pool with online self-update","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"searched not found","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall|attention","persistence":"cross-session","update":"parametric-edit","unit":"weight","governance":"opaque","conflict":"n/a","description":"Augments a transformer with a fixed-size 1B-parameter memory pool distributed across all layers; new knowledge is injected by updating the pool parameters during inference, not via context extension. Distinct from M+ (already in landscape) which extends with scalable long-term storage — this is the original fixed-capacity pool.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memoryllm-original--arxiv-2402-04624"},{"id":"memos-memtensor--gh-memtensor-memos","name":"MemOS (MemTensor)","url":"https://github.com/MemTensor/MemOS","section":"Research / specialised systems","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"\"Memory operating system\" + MemCubes","license":"Apache-2.0","latest_release":"v2.0.14 2026-04-23","deployment":"searched not found","pricing":"searched not found","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"8.9k★ +113/mo TypeScript","storage":"hybrid","retrieval":"similarity|hybrid-rerank","persistence":"long-term","update":"consolidation","unit":"chunk","governance":"inspectable","conflict":"n/a","description":"Treats memory as an OS-managed resource with explicit allocation and process-like scoping. MemCubes unify parametric, activation, and plaintext memory.","cited_cells":25,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memos-memtensor--gh-memtensor-memos"},{"id":"memp--arxiv-2508-06433","name":"Memp","url":"https://arxiv.org/abs/2508.06433","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Agent procedural memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"cross-session","update":"extraction","unit":"skill","governance":"n/a","conflict":"llm-arbitrate","description":"Explores agent procedural memory.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memp--arxiv-2508-06433"},{"id":"mempalace--mempalace-net","name":"MemPalace","url":"https://www.mempalace.net/","section":"Research / specialised systems","tier":2,"status":"active","last_verified_at":"2026-10-28","type":"Verbatim chunks + spatial metaphor","license":"MIT","latest_release":"v3.3.4 2026-05-01","deployment":"searched not found","pricing":"searched not found","mcp_support":"no first-party MCP adapter published as of 2026-05; community connectors may exist.","a2a_support":"no Google A2A (Agent2Agent) integration documented as of 2026-05.","backend_storage":"searched not found","llm_lock":"","github":"51.3k★ +49.6k/mo Python","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"append-only","unit":"chunk","governance":"inspectable","conflict":"append","description":"Spatial-metaphor memory (wings / rooms / halls) on top of verbatim chunked storage. Independent analysis suggests the score is driven by verbatim storage + ChromaDB defaults rather than the palace structure.","cited_cells":27,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/mempalace--mempalace-net"},{"id":"memr--arxiv-2512-20237","name":"MemR³","url":"https://arxiv.org/abs/2512.20237","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Closed-loop retrieve / reflect / answer router","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity|agentic","persistence":"long-term","update":"read-only","unit":"chunk","governance":"inspectable","conflict":"n/a","description":"Closed-loop retrieval controller replacing single-shot retrieve-then-answer. Iterative router chooses between retrieve, reflect, answer; global evidence-gap tracker makes retrieval reasoning explicit and auditable.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memr--arxiv-2512-20237"},{"id":"memreasoner--openreview-net","name":"MemReasoner","url":"https://openreview.net/forum?id=ODcMy97cVZ","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-05-14","type":"Iterative bidirectional-GRU re-read","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv-cache","retrieval":"attention","persistence":"session","update":"overwrite","unit":"kv-token","governance":"n/a","conflict":"n/a","description":"IBM Research. Memory-augmented LLM architecture for multi-hop reasoning. Iteratively re-reads a temporal latent memory with bidirectional GRUs until readout stabilises. NeurIPS 2024.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memreasoner--openreview-net"},{"id":"memrl--arxiv-2601-03192","name":"MemRL","url":"https://arxiv.org/abs/2601.03192","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Runtime RL on episodic memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"agent-controlled","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Self-evolving agents via runtime reinforcement learning on episodic memory.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memrl--arxiv-2601-03192"},{"id":"memsearcher--arxiv-2511-02805","name":"MemSearcher","url":"https://arxiv.org/abs/2511.02805","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"RL memory + search + reasoning","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity|agentic","persistence":"long-term","update":"agent-controlled","unit":"chunk","governance":"n/a","conflict":"llm-arbitrate","description":"Trains LLMs to reason, search, and manage memory via RL.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memsearcher--arxiv-2511-02805"},{"id":"memtool--arxiv-2507-21428","name":"MemTool","url":"https://arxiv.org/abs/2507.21428","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Short-term memory for tool calling","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv","retrieval":"exact-match","persistence":"session","update":"agent-controlled","unit":"kv-token","governance":"inspectable","conflict":"llm-arbitrate","description":"Optimises short-term memory management for dynamic tool calling.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memtool--arxiv-2507-21428"},{"id":"memtree--arxiv-2410-14052","name":"MemTree","url":"https://arxiv.org/abs/2410.14052","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Hierarchical tree memory for LLM agents","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"vector","retrieval":"similarity|graph-traversal","persistence":"long-term","update":"extraction","unit":"fact","governance":"opaque","conflict":"n/a","description":"Builds an LLM-driven hierarchical tree of facts; queries traverse the tree from root to leaves, balancing recall coverage with depth.","cited_cells":5,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memtree--arxiv-2410-14052"},{"id":"memverse--arxiv-2512-03627","name":"MemVerse","url":"https://arxiv.org/abs/2512.03627","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Multimodal lifelong learning","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity","persistence":"lifelong","update":"consolidation","unit":"episode","governance":"n/a","conflict":"llm-arbitrate","description":"Multimodal memory system for lifelong learning agents.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/memverse--arxiv-2512-03627"},{"id":"merge--arxiv-2403-20327","name":"MeRGE","url":"https://arxiv.org/abs/2403.20327","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Memory merging via embeddings","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"consolidation","unit":"fact","governance":"opaque","conflict":"n/a","description":"Embedding-space memory consolidation — merges semantically similar memories without LLM arbitration to reduce storage cost.","cited_cells":5,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/merge--arxiv-2403-20327"},{"id":"metagpt--arxiv-2308-00352","name":"MetaGPT","url":"https://arxiv.org/abs/2308.00352","section":"Recent method papers — theorized, no distinct product","tier":2,"status":"active","last_verified_at":"2026-10-23","type":"Multi-agent collaborative framework","license":"MIT","latest_release":"v0.8.1 (released March 2 2025); v0.8.2 current","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"file + pluggable","llm_lock":"","github":"67.5k GitHub stars (FoundationAgents/MetaGPT)","storage":"file|vector","retrieval":"similarity|injection","persistence":"cross-session","update":"overwrite|extraction","unit":"document|episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Meta-programming for a multi-agent collaborative framework.","cited_cells":30,"third_party_cells":4,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/metagpt--arxiv-2308-00352"},{"id":"mia-memory-intelligence-agent--arxiv-2604-04503","name":"MIA — Memory Intelligence Agent","url":"https://arxiv.org/abs/2604.04503","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Manager-Planner-Executor with bidirectional parametric/non-parametric loop","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"parametric-recall|similarity","persistence":"long-term","update":"consolidation","unit":"trajectory","governance":"n/a","conflict":"llm-arbitrate","description":"Manager stores compressed non-parametric search trajectories; Planner is a parametric memory agent producing search plans; Executor acts on plans. Planner evolves via test-time learning (on-the-fly weight updates during inference), creating a bidirectional conversion loop between parametric and non-parametric memory.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/mia-memory-intelligence-agent--arxiv-2604-04503"},{"id":"mind2web--osu-nlp-group-github-io","name":"Mind2Web","url":"https://osu-nlp-group.github.io/Mind2Web/","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Benchmark","license":"MIT (code); dataset CC-BY-4.0","latest_release":"last commit 2025-11-05","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/OSU-NLP-Group/Mind2Web (1036★)","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Dataset and benchmark for generalist web agents with 2,350 language-instruction tasks on 137 real websites across 31 domains.","cited_cells":12,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/mind2web--osu-nlp-group-github-io"},{"id":"minedojo--minedojo-org","name":"MineDojo","url":"https://minedojo.org/","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Open-ended Minecraft agent benchmark + knowledge base","license":"MIT (env wrapper); Mojang EULA on Minecraft","latest_release":"","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"2.4k★ Python","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"NVIDIA + Caltech open-ended embodied-agent benchmark on Minecraft. 3000+ programmatic + creative tasks across survival / harvest / tech-tree / combat / creative. Includes an internet-scale knowledge base (730K YouTube videos, 7K wiki pages, 340K Reddit posts) for grounding. Released at NeurIPS 2022 (outstanding paper award).","cited_cells":28,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/minedojo--minedojo-org"},{"id":"minerl-diamond-challenge--minerl-io","name":"MineRL Diamond Challenge","url":"https://minerl.io/","section":"Memory benchmarks & evaluation","tier":2,"status":"active","last_verified_at":"2026-05-14","type":"Minecraft RL benchmark — obtain a diamond","license":"MIT","latest_release":"","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"~700★ Python","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"MineRL is a research project / competition series using Minecraft as a long-horizon RL environment. The flagship Diamond Challenge tasks an agent with obtaining a diamond from raw world — requires deep tech-tree planning across thousands of steps. DreamerV3 (DeepMind, 2023-12) famously became the first to solve it from scratch without human data.","cited_cells":28,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/minerl-diamond-challenge--minerl-io"},{"id":"mirix--arxiv-2507-07957","name":"MIRIX","url":"https://arxiv.org/abs/2507.07957","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Multi-agent memory system","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity|graph-traversal","persistence":"long-term","update":"extraction","unit":"episode","governance":"user-controllable","conflict":"llm-arbitrate","description":"Multi-agent memory system for LLM-based agents.","cited_cells":16,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/mirix--arxiv-2507-07957"},{"id":"mlp-memory--arxiv-2508-01832","name":"MLP Memory","url":"https://arxiv.org/abs/2508.01832","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Retriever-pretrained external memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"read-only","unit":"document","governance":"n/a","conflict":"n/a","description":"Language modeling with retriever-pretrained external memory.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/mlp-memory--arxiv-2508-01832"},{"id":"mmag--arxiv-2512-01710","name":"MMAG","url":"https://arxiv.org/abs/2512.01710","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Mixed memory-augmented generation","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity|hybrid-rerank","persistence":"long-term","update":"consolidation","unit":"chunk","governance":"user-controllable","conflict":"llm-arbitrate","description":"Mixed memory-augmented generation for LLM applications.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/mmag--arxiv-2512-01710"},{"id":"mot-memory-of-thought--arxiv-2305-05181","name":"MoT (Memory-of-Thought)","url":"https://arxiv.org/abs/2305.05181","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"research-code","last_verified_at":"2026-10-28","type":"Pre-test self-reflection cache","license":"no-license","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"24★ Python","storage":"vector","retrieval":"similarity","persistence":"session","update":"append-only","unit":"chunk","governance":"n/a","conflict":"llm-arbitrate","description":"Before test time, LLM reasons over unlabeled data and stores high-confidence intermediate thoughts as external memory. At test time, retrieves relevant stored thoughts to augment reasoning. No annotated data, no parameter updates.","cited_cells":25,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/mot-memory-of-thought--arxiv-2305-05181"},{"id":"mpo","name":"MPO","url":"","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Meta plan optimisation","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"cross-session","update":"extraction","unit":"skill","governance":"n/a","conflict":"llm-arbitrate","description":"Boosts LLM agents with meta plan optimisation. EMNLP 2025.","cited_cells":16,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/mpo"},{"id":"namms--arxiv-2410-13166","name":"NAMMs","url":"https://arxiv.org/abs/2410.13166","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"research-code","last_verified_at":"2026-10-28","type":"Evolution-trained KV-cache memory","license":"Unlicensed","latest_release":"no releases","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"357★ Python","storage":"kv-cache","retrieval":"attention","persistence":"session","update":"evict-oldest","unit":"kv-token","governance":"inspectable","conflict":"n/a","description":"Sakana AI. \"An Evolved Universal Transformer Memory.\" Neural Attention Memory Models trained via evolution (overcomes non-differentiability of token-keep/discard).","cited_cells":25,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/namms--arxiv-2410-13166"},{"id":"naver-provence-pisco--arxiv-2501-16214","name":"NAVER Provence + PISCO","url":"https://arxiv.org/abs/2501.16214","section":"Recent method papers — theorized, no distinct product","tier":2,"status":"active","last_verified_at":"2026-10-28","type":"RAG context pruning + soft compression","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"searched not found","llm_lock":"","github":"","storage":"kv-cache","retrieval":"cache-lookup|attention","persistence":"session","update":"evict-oldest","unit":"kv-token","governance":"opaque","conflict":"n/a","description":"Two complementary NAVER Labs Europe papers (Jan 2025). Provence fine-tunes DeBERTa as a sequence labeler to dynamically detect needed pruning amount per query and mask irrelevant context spans before generation, unifying pruning with reranking in one pass. PISCO is a distillation-only approach (no annotated data, no pretraining) that trains an LLM to decode from a 16x-compressed document representation.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/naver-provence-pisco--arxiv-2501-16214"},{"id":"nemori--arxiv-2508-03341","name":"Nemori","url":"https://arxiv.org/abs/2508.03341","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Self-organising cognitive-inspired","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity","persistence":"long-term","update":"consolidation","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Self-organising agent memory inspired by cognitive science.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/nemori--arxiv-2508-03341"},{"id":"nethack-learning-environment-nle--gh-heiner-nle","name":"NetHack Learning Environment (NLE)","url":"https://github.com/heiner/nle","section":"Memory benchmarks & evaluation","tier":2,"status":"active","last_verified_at":"2026-05-14","type":"Roguelike RL benchmark — NetHack as agent environment","license":"NetHack General Public License","latest_release":"v0.9.x series (active as of 2024)","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"1.0k★ Python/C","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"DeepMind / FAIR's NetHack Learning Environment — a Gym-compatible interface to NetHack, the canonical roguelike. Massive procedural state space, permadeath, deep item / skill interactions, ASCII partial-observability. Released at NeurIPS 2020; canonical long-horizon partial-observability RL benchmark.","cited_cells":30,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/nethack-learning-environment-nle--gh-heiner-nle"},{"id":"neural-episodic-control--arxiv-1703-01988","name":"Neural Episodic Control","url":"https://arxiv.org/abs/1703.01988","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"research-code","last_verified_at":"2026-10-23","type":"Differentiable episodic dictionary","license":"no-license","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"59★ Python","storage":"kv","retrieval":"similarity","persistence":"long-term","update":"overwrite","unit":"episode","governance":"inspectable","conflict":"n/a","description":"Pritzel et al. Agent stores past state-action-value tuples in a differentiable dictionary; recall via approximate nearest neighbour. ICML 2017.","cited_cells":22,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/neural-episodic-control--arxiv-1703-01988"},{"id":"niah-needle-in-a-haystack--gh-gkamradt-llmtest-needleinahaystack","name":"NIAH (Needle in a Haystack)","url":"https://github.com/gkamradt/LLMTest_NeedleInAHaystack","section":"Memory benchmarks & evaluation","tier":4,"status":"active","last_verified_at":"2026-05-14","type":"Single-fact verbatim retrieval from long context","license":"MIT","latest_release":"no releases","deployment":"not applicable — run locally","pricing":"not applicable — open source; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"2.3k★ Jupyter Notebook","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Places a known sentence (\"needle\") at varying positions inside a long irrelevant document (\"haystack\") and prompts the model to retrieve it. Tests context lengths from 2K to model max. Greg Kamradt, 2023, no formal venue.","cited_cells":25,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/niah-needle-in-a-haystack--gh-gkamradt-llmtest-needleinahaystack"},{"id":"nl2gensym--arxiv-2510-09355","name":"NL2GenSym","url":"https://arxiv.org/abs/2510.09355","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Soar production-rule generation + Execution-Grounded Critic","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv","retrieval":"exact-match","persistence":"long-term","update":"extraction","unit":"skill","governance":"deterministic","conflict":"llm-arbitrate","description":"First end-to-end pipeline translating natural language into executable Soar rules. Imports the full Soar production-rule cycle: working memory, operator proposal/application, chunking. LLM generator proposes new productions; critic validates via live Soar execution.","cited_cells":16,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/nl2gensym--arxiv-2510-09355"},{"id":"nvidia-nemotron-3--arxiv-2512-20856","name":"NVIDIA Nemotron 3","url":"https://arxiv.org/abs/2512.20856","section":"Recent method papers — theorized, no distinct product","tier":1,"status":"active","last_verified_at":"2026-10-28","type":"Hybrid Mamba + sparse KV-cache, 1M-token context","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"searched not found","llm_lock":"","github":"","storage":"kv-cache|parametric","retrieval":"attention|cache-lookup","persistence":"session","update":"read-only","unit":"kv-token","governance":"opaque","conflict":"n/a","description":"Interleaves sparse MoE layers with Mamba-2 layers; Mamba layers maintain a fixed-size recurrent state (constant memory during generation) rather than a growing KV cache. Attention layers use GQA with only 2 KV heads. Two parallel memory systems: explicit small KV cache (attention layers only) and implicit compressed history in the Mamba state. Post-trained via multi-environment RL for multi-step tool use.","cited_cells":29,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/nvidia-nemotron-3--arxiv-2512-20856"},{"id":"oasis--arxiv-2411-11581","name":"OASIS","url":"https://arxiv.org/abs/2411.11581","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Million-agent social simulation","license":"Apache-2.0","latest_release":"v0.2.5 2025-12-04","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"4.5k★ +600/mo Python","storage":"hybrid","retrieval":"similarity","persistence":"session","update":"extraction","unit":"profile","governance":"n/a","conflict":"llm-arbitrate","description":"Open agent social interaction simulations with one million agents.","cited_cells":25,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/oasis--arxiv-2411-11581"},{"id":"openai-gym-retro--gh-openai-retro","name":"OpenAI Gym Retro","url":"https://github.com/openai/retro","section":"Memory benchmarks & evaluation","tier":3,"status":"archived","last_verified_at":"2026-10-28","type":"Retro arcade RL benchmark — Sega / NES / SNES game integration","license":"MIT","latest_release":"","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"3.5k★ Python/C","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"OpenAI's Gym Retro — wraps thousands of classic arcade games (Sega Genesis, NES, SNES, Atari 2600, Game Boy) as Gym RL environments via libretro emulators. Released 2018; powered the OpenAI Retro Contest where agents had to generalise from train-set games to held-out levels of Sonic. Largely supplanted by Atari 100k / Procgen / etc. but still the most-popular path for adding arbitrary retro games to RL pipelines.","cited_cells":30,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/openai-gym-retro--gh-openai-retro"},{"id":"osrs-bench-old-school-runescape-agent-benchmark--gh-grahamannett-osrs-bench","name":"OSRS Bench (Old School RuneScape agent benchmark)","url":"https://github.com/grahamannett/osrs-bench","section":"Memory benchmarks & evaluation","tier":4,"status":"inactive","last_verified_at":"2026-05-14","type":"Long-horizon MMO agent benchmark — Old School RuneScape","license":"MIT (assumed — community repo, no explicit LICENSE file checked)","latest_release":"","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"small community repo (<50★), Python","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Community benchmark suite for LLM agents playing Old School RuneScape, a sandbox MMORPG with deep partial-observability, skill-grinding loops, and economy interaction. Tests long-horizon planning (hour-to-day quests), tool inventory management, and visual UI comprehension. Community-built; not affiliated with Jagex.","cited_cells":24,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/osrs-bench-old-school-runescape-agent-benchmark--gh-grahamannett-osrs-bench"},{"id":"osworld--os-world-github-io","name":"OSWorld","url":"https://os-world.github.io/","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Desktop OS agent benchmark — 369 tasks across Ubuntu / Windows / macOS","license":"Apache-2.0","latest_release":"","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"1.7k★ Python","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"HKU NLP + Salesforce's OSWorld — full-desktop agent benchmark with 369 tasks spanning real applications (LibreOffice, GIMP, Chrome, VS Code, Thunderbird, Files) across Ubuntu, Windows, and macOS. Tasks require multi-application workflows; evaluation via post-task state inspection (not just screenshot match). Released NeurIPS 2024.","cited_cells":28,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/osworld--os-world-github-io"},{"id":"planning-from-imagination--arxiv-2412-01857","name":"Planning from Imagination","url":"https://arxiv.org/abs/2412.01857","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Episodic simulation memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"session","update":"read-only","unit":"trajectory","governance":"n/a","conflict":"llm-arbitrate","description":"Episodic simulation and memory for vision-language navigation.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/planning-from-imagination--arxiv-2412-01857"},{"id":"pokemon-red-benchmark-speedrun-completion--gh-pwhiddy-pokemonredexperiments","name":"Pokemon Red benchmark (speedrun / completion)","url":"https://github.com/PWhiddy/PokemonRedExperiments","section":"Memory benchmarks & evaluation","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Pokemon Red completion benchmark — long-horizon JRPG agent task","license":"MIT (Whiddy repo)","latest_release":"","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"7.6k★ Python (PokemonRedExperiments)","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Pokemon Red has emerged as a community + academic benchmark for long-horizon agentic AI. Notable instances: Peter Whiddy's PokemonRedExperiments (RL via PPO + curiosity, viral YouTube series 2023); Anthropic's ClaudePlaysPokemon Twitch stream (2025-02); Google's GeminiPlaysPokemon (2025); MorphLabs PokemonRedRL. The benchmark protocol varies: badge count, gym progression, full Elite-Four completion. Common across all: ~5k-20k+ step horizon, sparse rewards, deep partial-observability.","cited_cells":30,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/pokemon-red-benchmark-speedrun-completion--gh-pwhiddy-pokemonredexperiments"},{"id":"pose--arxiv-2309-10400","name":"PoSE","url":"https://arxiv.org/abs/2309.10400","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-23","type":"Positional Skip-wisE training","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"kv-cache","retrieval":"attention","persistence":"ephemeral","update":"append-only","unit":"kv-token","governance":"opaque","conflict":"out-of-scope","description":"Trains long-context models by simulating long positions with skip-wise positional encodings — avoids high training cost of true long sequences.","cited_cells":5,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/pose--arxiv-2309-10400"},{"id":"prime--arxiv-2509-22315","name":"PRIME","url":"https://arxiv.org/abs/2509.22315","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Planning + retrieval-integrated memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity|agentic","persistence":"long-term","update":"consolidation","unit":"chunk","governance":"n/a","conflict":"llm-arbitrate","description":"Planning- and retrieval-integrated memory for enhanced reasoning.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/prime--arxiv-2509-22315"},{"id":"principles--arxiv-2509-17459","name":"PRINCIPLES","url":"https://arxiv.org/abs/2509.17459","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Synthetic strategy memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"extraction","unit":"skill","governance":"n/a","conflict":"llm-arbitrate","description":"Synthetic strategy memory for proactive dialogue agents.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/principles--arxiv-2509-17459"},{"id":"procgen-benchmark-suite--gh-openai-procgen","name":"PROCGEN benchmark suite","url":"https://github.com/openai/procgen","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Procedurally-generated RL game suite — generalisation eval","license":"MIT","latest_release":"","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"980★ Python/C","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"OpenAI's 16-game procedurally-generated benchmark for evaluating RL generalisation. Unlike Atari (fixed levels memorisable), every PROCGEN episode uses a different procedural seed → test-set generalisation is the headline metric. Released at ICML 2020. Standard benchmark for studying generalisation, exploration, transfer.","cited_cells":30,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/procgen-benchmark-suite--gh-openai-procgen"},{"id":"pwm-policy-learning-with-large-world-models--arxiv-2407-02466","name":"PWM (Policy Learning with Large World Models)","url":"https://arxiv.org/abs/2407.02466","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Multi-task world-model + first-order gradient policies","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"trajectory","governance":"inspectable","conflict":"n/a","description":"Pre-trains a large multi-task world model on offline data across many continuous-control tasks, then extracts task-specific policies by differentiating through the model with first-order gradient optimisation. World model serves as compressed environment memory enabling policy transfer.","cited_cells":18,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/pwm-policy-learning-with-large-world-models--arxiv-2407-02466"},{"id":"r-mem--arxiv-2502-15957","name":"R³Mem","url":"https://arxiv.org/abs/2502.15957","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Reversible compression memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity","persistence":"long-term","update":"consolidation","unit":"summary","governance":"n/a","conflict":"llm-arbitrate","description":"Bridging memory retention and retrieval via reversible compression.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/r-mem--arxiv-2502-15957"},{"id":"r2i--arxiv-2403-04253","name":"R2I","url":"https://arxiv.org/abs/2403.04253","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"research-code","last_verified_at":"2026-10-28","type":"DreamerV3 + S4 long-range memory","license":"MIT","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"81★ Python","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"trajectory","governance":"n/a","conflict":"n/a","description":"Memory-enhanced model-based RL agent built on DreamerV3. Modifies S4 to handle long temporal dependencies. Keeps DreamerV3's generality.","cited_cells":24,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/r2i--arxiv-2403-04253"},{"id":"razorattention--openreview-net","name":"RazorAttention","url":"https://openreview.net/forum?id=tkiZQlL04w","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"KV-cache compression via retrieval heads","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv-cache","retrieval":"attention|cache-lookup","persistence":"session","update":"evict-oldest","unit":"kv-token","governance":"n/a","conflict":"n/a","description":"Efficient KV-cache compression through retrieval heads.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/razorattention--openreview-net"},{"id":"readagent--arxiv-2402-09727","name":"ReadAgent","url":"https://arxiv.org/abs/2402.09727","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Gist-episodic compression + on-demand lookup","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv","retrieval":"similarity|extraction-pull","persistence":"session","update":"read-only","unit":"summary","governance":"n/a","conflict":"n/a","description":"Google Research / DeepMind. Prompts LLM to segment long documents into \"pages,\" compresses each to a short gist stored as episodic memory, re-reads original pages on demand at answer time. Pure prompting, no fine-tuning.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/readagent--arxiv-2402-09727"},{"id":"reasoningbank--arxiv-2509-25140","name":"ReasoningBank","url":"https://arxiv.org/abs/2509.25140","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Reasoning-memory self-evolution","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"extraction","unit":"trajectory","governance":"inspectable","conflict":"llm-arbitrate","description":"Scales agent self-evolution with reasoning memory.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/reasoningbank--arxiv-2509-25140"},{"id":"recurrent-memory-transformer-rmt--arxiv-2207-06881","name":"Recurrent Memory Transformer (RMT)","url":"https://arxiv.org/abs/2207.06881","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"research-code","last_verified_at":"2026-10-23","type":"Segment-level recurrence + memory tokens","license":"Unlicensed","latest_release":"no releases","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"779★ Jupyter Notebook","storage":"kv-cache","retrieval":"attention","persistence":"session","update":"overwrite","unit":"kv-token","governance":"n/a","conflict":"n/a","description":"Bulatov / Kuratov / Burtsev. Adds memory tokens passed between segments — no architectural change to the Transformer.","cited_cells":25,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/recurrent-memory-transformer-rmt--arxiv-2207-06881"},{"id":"recurrentgpt--arxiv-2305-13304","name":"RecurrentGPT","url":"https://arxiv.org/abs/2305.13304","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"research-code","last_verified_at":"2026-10-28","type":"Long-text interactive generation","license":"GPL-3.0","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"1.0k★ Python","storage":"hybrid","retrieval":"similarity|injection","persistence":"session","update":"overwrite","unit":"summary","governance":"editable","conflict":"append","description":"Interactive generation of arbitrarily long text.","cited_cells":25,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/recurrentgpt--arxiv-2305-13304"},{"id":"reflexion--arxiv-2303-11366","name":"Reflexion","url":"https://arxiv.org/abs/2303.11366","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Verbal reinforcement learning","license":"MIT","latest_release":"no releases","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"3.1k★ Python","storage":"kv","retrieval":"injection","persistence":"session","update":"append-only","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Language agents with verbal reinforcement learning.","cited_cells":26,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/reflexion--arxiv-2303-11366"},{"id":"repairagent--arxiv-2403-17134","name":"RepairAgent","url":"https://arxiv.org/abs/2403.17134","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Autonomous program repair agent","license":"no data — searched arxiv abstract; no specific value reported","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"92★ Python","storage":"kv","retrieval":"agentic","persistence":"session","update":"agent-controlled","unit":"episode","governance":"n/a","conflict":"llm-arbitrate","description":"Autonomous LLM-based agent for program repair.","cited_cells":22,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/repairagent--arxiv-2403-17134"},{"id":"resum--arxiv-2509-13313","name":"ReSum","url":"https://arxiv.org/abs/2509.13313","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Long-horizon search via summarisation","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv","retrieval":"similarity","persistence":"session","update":"consolidation","unit":"summary","governance":"n/a","conflict":"llm-arbitrate","description":"Unlocks long-horizon search intelligence via context summarisation.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/resum--arxiv-2509-13313"},{"id":"ret-llm--arxiv-2305-14322","name":"RET-LLM","url":"https://arxiv.org/abs/2305.14322","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"General read-write memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv","retrieval":"exact-match|similarity","persistence":"long-term","update":"overwrite","unit":"fact","governance":"inspectable","conflict":"overwrite","description":"General read-write memory for large language models.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/ret-llm--arxiv-2305-14322"},{"id":"retroformer--arxiv-2308-02151","name":"Retroformer","url":"https://arxiv.org/abs/2308.02151","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"research-code","last_verified_at":"2026-10-28","type":"Retrospective policy-gradient agents","license":"Apache-2.0","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"39★ Python","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"episode","governance":"n/a","conflict":"llm-arbitrate","description":"Retrospective LLM agents with policy-gradient optimisation.","cited_cells":24,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/retroformer--arxiv-2308-02151"},{"id":"rgmem--arxiv-2510-16392","name":"RGMem","url":"https://arxiv.org/abs/2510.16392","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Renormalisation-group profile evolution","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"profile","governance":"inspectable","conflict":"llm-arbitrate","description":"Renormalization-group-based memory evolution for user profiles.","cited_cells":20,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/rgmem--arxiv-2510-16392"},{"id":"riken-tensor-decomposition-incremental-learning--icml-cc","name":"RIKEN Tensor Decomposition Incremental Learning","url":"https://icml.cc/virtual/2025/poster/44196","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Tensor-decomposed exemplar compression","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"episode","governance":"n/a","conflict":"n/a","description":"Uses tensor decomposition to exploit low intrinsic dimensionality and pixel correlation of stored exemplar images, achieving high compression while preserving discriminative information for class-incremental learning. Compresses the replay buffer used in catastrophic-forgetting prevention. RIKEN AIP + Guangdong U. of Technology.","cited_cells":16,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/riken-tensor-decomposition-incremental-learning--icml-cc"},{"id":"rl-based-continual-learning-as-memory--cameronrwolfe-substack-com","name":"RL-Based Continual Learning as Memory","url":"https://cameronrwolfe.substack.com/p/rl-continual-learning","section":"Theoretical / informal — ideas without a paper","tier":5,"status":"active","last_verified_at":"2026-10-23","type":"Memory-in-weights via on-policy RL","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"weight","governance":"n/a","conflict":"n/a","description":"On-policy RL is naturally more resistant to catastrophic forgetting than supervised fine-tuning — its online nature biases learning toward low-distribution-shift updates. Proposal: RL post-training as a viable mechanism for agents to accumulate long-term memory in weights , not just in an external store.","cited_cells":11,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/rl-based-continual-learning-as-memory--cameronrwolfe-substack-com"},{"id":"rl-developer-memory--arxiv-2605-01567","name":"RL Developer Memory","url":"https://arxiv.org/abs/2605.01567","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Safety-gated MCP architecture for RL coding-agent memory","license":"","latest_release":"","deployment":"Local-first / self-hosted MCP server (paper architecture)","pricing":"","mcp_support":"native (first-party MCP design)","a2a_support":"not supported","backend_storage":"local-first (MCP-native)","llm_lock":"","github":"searched not found","storage":"kv","retrieval":"similarity","persistence":"long-term","update":"reward-gated","unit":"episode","governance":"user-controllable","conflict":"deterministic-ranker","description":"Local-first MCP architecture for RL coding agents that treats memory selection as a logged contextual decision process. Three MCP tools: issue_match (ranks candidates + records telemetry), issue_feedback (maps raw labels to bounded rewards), issue_record_resolution (links resolutions to retrieval events). Deterministic ranker in production while a contextual-bandit residual policy runs in shadow mode behind off-policy-evaluation gates.","cited_cells":31,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/rl-developer-memory--arxiv-2605-01567"},{"id":"rmm-reflective-memory-management--arxiv-2503-08026","name":"RMM (Reflective Memory Management)","url":"https://arxiv.org/abs/2503.08026","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Prospective + retrospective dual reflection","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity|hybrid-rerank","persistence":"cross-session","update":"consolidation","unit":"summary","governance":"n/a","conflict":"llm-arbitrate","description":"Combines forward-looking Prospective Reflection (multi-granularity summarisation into personalised memory bank) with backward-looking Retrospective Reflection (RL-driven retrieval refinement guided by cited evidence). Operates at utterance / turn / session scopes.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/rmm-reflective-memory-management--arxiv-2503-08026"},{"id":"ruler--arxiv-2404-06654","name":"RULER","url":"https://arxiv.org/abs/2404.06654","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Configurable multi-task long-context stress test","license":"Apache-2.0","latest_release":"no releases","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"1.5k★ Python","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"NVIDIA. Extends NIAH with multi-needle retrieval, multi-hop entity tracing, aggregation, QA — configurable for arbitrary length / difficulty. Evaluated 17 long-context models on 13 tasks.","cited_cells":29,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/ruler--arxiv-2404-06654"},{"id":"rwkv-7--gh-blinkdl-rwkv-lm","name":"RWKV-7","url":"https://github.com/BlinkDL/RWKV-LM","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"RNN with transformer-level perf","license":"Apache-2.0","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"","github":"searched not found","storage":"cache","retrieval":"attention","persistence":"session","update":"overwrite|evict-oldest","unit":"summary|token","governance":"opaque","conflict":"out-of-scope","description":"RWKV recurrent-LM family — linear-time, constant-memory inference; v7 ('Goose') is the SotA recurrent LM. Community-driven OSS, Linux Foundation hosted.","cited_cells":9,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/rwkv-7--gh-blinkdl-rwkv-lm"},{"id":"s--arxiv-2307-14984","name":"S³","url":"https://arxiv.org/abs/2307.14984","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"LLM social-network simulation","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"session","update":"extraction","unit":"profile","governance":"n/a","conflict":"llm-arbitrate","description":"Social-network simulation system with LLM-empowered agents.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/s--arxiv-2307-14984"},{"id":"sage","name":"SAGE","url":"","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Reflective memory-augmented agent","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"extraction","unit":"episode","governance":"n/a","conflict":"llm-arbitrate","description":"Self-evolving agents with reflective and memory-augmented abilities. Neurocomputing 2025 .","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/sage"},{"id":"sakana-ctm--arxiv-2505-05522","name":"Sakana CTM","url":"https://arxiv.org/abs/2505.05522","section":"Recent method papers — theorized, no distinct product","tier":2,"status":"active","last_verified_at":"2026-10-23","type":"Per-neuron history + neural synchronization as latent","license":"Apache-2.0","latest_release":"no releases","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"searched not found","llm_lock":"","github":"1.8k★ Python","storage":"parametric","retrieval":"parametric-recall|attention","persistence":"session","update":"read-only","unit":"kv-token","governance":"opaque","conflict":"n/a","description":"Continuous Thought Machines. Each neuron maintains its own history of incoming activations via private per-neuron weight parameters; the model operates along a self-generated \"thought step\" timeline rather than the input sequence timeline. Neural synchronization across neurons is used directly as a latent representation. Distinct from Sakana NAMM (evolutionary attention memory) — CTM is an architectural rethink of the recurrent state itself.","cited_cells":21,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/sakana-ctm--arxiv-2505-05522"},{"id":"scm--arxiv-2304-13343","name":"SCM","url":"https://arxiv.org/abs/2304.13343","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"research-code","last_verified_at":"2026-10-28","type":"Self-controlled memory framework","license":"MIT","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"50★ Jupyter Notebook","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"agent-controlled","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Enhances LLMs with self-controlled memory framework.","cited_cells":24,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/scm--arxiv-2304-13343"},{"id":"sculptor--arxiv-2508-04664","name":"Sculptor","url":"https://arxiv.org/abs/2508.04664","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Active context management","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv","retrieval":"agentic","persistence":"session","update":"agent-controlled","unit":"chunk","governance":"agent-controlled","conflict":"llm-arbitrate","description":"Empowers LLMs with cognitive agency via active context management.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/sculptor--arxiv-2508-04664"},{"id":"sdm-activations--arxiv-2509-12760","name":"SDM Activations","url":"https://arxiv.org/abs/2509.12760","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Similarity-Distance-Magnitude activations","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"attention","persistence":"session","update":"read-only","unit":"kv-token","governance":"inspectable","conflict":"n/a","description":"Novel activation approach for memory.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/sdm-activations--arxiv-2509-12760"},{"id":"seagent--arxiv-2508-04700","name":"SEAgent","url":"https://arxiv.org/abs/2508.04700","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Self-evolving computer-use agent","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity|agentic","persistence":"long-term","update":"agent-controlled","unit":"skill","governance":"inspectable","conflict":"llm-arbitrate","description":"Self-evolving computer-use agent with autonomous learning.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/seagent--arxiv-2508-04700"},{"id":"secom--openreview-net","name":"SeCom","url":"https://openreview.net/forum?id=xKDZAW0He3","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Personalised memory construction + retrieval","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"cross-session","update":"consolidation","unit":"summary","governance":"n/a","conflict":"llm-arbitrate","description":"Memory construction and retrieval for personalised conversational agents.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/secom--openreview-net"},{"id":"self-param--arxiv-2410-00487","name":"SELF-PARAM","url":"https://arxiv.org/abs/2410.00487","section":"Recent method papers — theorized, no distinct product","tier":2,"status":"active","last_verified_at":"2026-10-28","type":"KL-distillation experience internalization","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"searched not found","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"weight","governance":"opaque","conflict":"n/a","description":"Encodes new experiences directly into the LLM's existing parameters by minimizing KL divergence between a teacher model (with context access) and a student model (without). Generates diverse QA pairs from new knowledge as the distillation signal. No new memory pool or external store added.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/self-param--arxiv-2410-00487"},{"id":"self-refine--arxiv-2303-17651","name":"Self-Refine","url":"https://arxiv.org/abs/2303.17651","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Iterative self-feedback","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"kv","retrieval":"extraction-pull","persistence":"session","update":"accretion","unit":"episode","governance":"plaintext","conflict":"n/a","description":"Iterative self-feedback loop — LLM critiques its own output and revises; in-context experiential memory of the current task.","cited_cells":6,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/self-refine--arxiv-2303-17651"},{"id":"sers--openreview-net","name":"SERS","url":"https://openreview.net/forum?id=jR1lvwexLt","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-05-14","type":"Self-evolving pseudo-rehearsal","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"episode","governance":"inspectable","conflict":"llm-arbitrate","description":"Dynamic regulariser driven by Wasserstein distance between task distributions; automatically relaxes or strengthens constraints in proportion to task similarity.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/sers--openreview-net"},{"id":"sgmem--arxiv-2509-21212","name":"SGMem","url":"https://arxiv.org/abs/2509.21212","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Sentence-graph memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"graph","retrieval":"graph-traversal","persistence":"long-term","update":"extraction","unit":"episode","governance":"n/a","conflict":"llm-arbitrate","description":"Sentence-graph memory for long-term conversational agents.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/sgmem--arxiv-2509-21212"},{"id":"shadowkv-bytedance--arxiv-2410-21465","name":"ShadowKV (ByteDance)","url":"https://arxiv.org/abs/2410.21465","section":"Recent method papers — theorized, no distinct product","tier":2,"status":"active","last_verified_at":"2026-10-28","type":"Low-rank CPU-offloaded KV cache + sparse reconstruction","license":"Apache-2.0","latest_release":"no releases","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"searched not found","llm_lock":"","github":"297★ Python","storage":"kv-cache","retrieval":"cache-lookup|attention","persistence":"session","update":"evict-oldest","unit":"kv-token","governance":"opaque","conflict":"n/a","description":"During prefill, value caches are offloaded to CPU and only low-rank pre-RoPE keys plus compressed landmarks and outliers are kept on GPU. During decoding, landmarks select chunk indices for key reconstruction and value fetching from CPU. Effectively unlimited context without GPU memory blowup. ICML 2025 Spotlight.","cited_cells":25,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/shadowkv-bytedance--arxiv-2410-21465"},{"id":"simplemem--arxiv-2601-02553","name":"SimpleMem","url":"https://arxiv.org/abs/2601.02553","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Paper","license":"MIT","latest_release":"last commit 2026-07-24","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/aiming-lab/SimpleMem (3844★)","storage":"column|vector","retrieval":"hybrid|agentic","persistence":"lifelong|long-term","update":"extraction|consolidation","unit":"fact","governance":"n/a","conflict":"n/a","description":"Efficient lifelong memory built from semantic structured compression into self-contained facts, online synthesis of related facts and intent-aware multi-index retrieval. Reports +26% F1 over Mem0 on LoCoMo with ~30x fewer tokens. ICML 2026; UNC/Berkeley/UCSC","cited_cells":13,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/simplemem--arxiv-2601-02553"},{"id":"skillweaver--arxiv-2504-07079","name":"SkillWeaver","url":"https://arxiv.org/abs/2504.07079","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Self-improving web agents","license":"MIT","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"121★ +2/mo Python","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"extraction","unit":"skill","governance":"n/a","conflict":"llm-arbitrate","description":"Web agents self-improve by discovering and honing skills.","cited_cells":23,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/skillweaver--arxiv-2504-07079"},{"id":"sleep-time-compute-letta--letta-com","name":"Sleep-time Compute (Letta)","url":"https://letta.com/blog/sleep-time-compute","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Paper","license":"Apache-2.0 (Letta Code)","latest_release":"Letta Code v0.34.30 (npm @letta-ai/letta-code, 2026-10-29)","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"letta-ai/letta-code — ~3.7k stars (current source; /sleeptime configures dreaming); letta-ai/letta ~25.3k stars, V1 server on its archive branch","storage":"kv","retrieval":"injection|agentic","persistence":"long-term","update":"consolidation|agent-controlled","unit":"summary|fact","governance":"inspectable","conflict":"llm-arbitrate","description":"Agents use idle time to reorganize and pre-reason over stored context ('learned context'). Letta ships this as sleep-time agents (v0.7.0+), where a background agent manages the primary agent's core memory. Traction: Shipped in Letta 0.7.0 and Letta Code (/sleeptime); arXiv 2504.13171. Letta's current source moved to letta-ai/letta-code; the V1 API server (where sleep-time agents first shipped) is retired to the letta-ai/letta archive branch. Letta Code's README now calls sleep-time compute 'dreaming': periodic dreaming via /sleeptime, manual /dream and /reflect, plus server-side dreaming.","cited_cells":12,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/sleep-time-compute-letta--letta-com"},{"id":"smartplay--gh-microsoft-smartplay","name":"SmartPlay","url":"https://github.com/Microsoft/SmartPlay","section":"Memory benchmarks & evaluation","tier":3,"status":"archived","last_verified_at":"2026-10-28","type":"LLM-agent benchmark across 6 games — Microsoft Research","license":"MIT","latest_release":"","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"170★ Python","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Microsoft Research's SmartPlay — LLM-agent benchmark on 6 games (Bandits, Rock-Paper-Scissors, Tower of Hanoi, MessengerEnv, Crafter, Minecraft). Tests 9 distinct capability axes including object understanding, planning, generalisation, error-handling, theory-of-mind. ICLR 2024. Bridges narrow-task benchmarks (MMLU) and full-agent benchmarks (AgentBench).","cited_cells":29,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/smartplay--gh-microsoft-smartplay"},{"id":"snapkv","name":"SnapKV","url":"","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"KV-cache pre-selection","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv-cache","retrieval":"attention|cache-lookup","persistence":"session","update":"evict-oldest","unit":"kv-token","governance":"n/a","conflict":"n/a","description":"LLM knows what you're seeking before generation. NeurIPS 2024.","cited_cells":21,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/snapkv"},{"id":"softcot--aclanthology-org","name":"SoftCoT","url":"https://aclanthology.org/2025.acl-long.1137/","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Soft chain of thought","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv-cache","retrieval":"attention","persistence":"session","update":"read-only","unit":"kv-token","governance":"inspectable","conflict":"n/a","description":"Soft chain-of-thought for efficient reasoning with LLMs. ACL 2025.","cited_cells":20,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/softcot--aclanthology-org"},{"id":"stale-cupmem--arxiv-2605-06527","name":"STALE (+ CUPMEM)","url":"https://arxiv.org/abs/2605.06527","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Benchmark","license":"unknown","latest_release":"searched not found","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"","storage":"none-trivial|kv","retrieval":"none|hybrid","persistence":"long-term","update":"read-only|consolidation","unit":"fact","governance":"n/a","conflict":"llm-arbitrate|out-of-scope","description":"Conflict/update benchmark of 400 scenarios with 1,200 queries (up to ~150K tokens) testing whether agents notice that later facts implicitly invalidate earlier memories; the best model scores 55.2%. The CUPMEM prototype adds write-time state adjudication. Traction: Directly targets implicit memory conflict; HKUST/Wuhan/CUHK","cited_cells":6,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/stale-cupmem--arxiv-2605-06527"},{"id":"streaming-llm--arxiv-2309-17453","name":"Streaming-LLM","url":"https://arxiv.org/abs/2309.17453","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Attention-sink streaming","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"kv","retrieval":"similarity","persistence":"session","update":"replacement","unit":"token","governance":"opaque","conflict":"n/a","description":"Attention-sink discovery — keep the first few tokens always cached; arbitrary-length inference with no fine-tuning. ICLR 2024.","cited_cells":6,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/streaming-llm--arxiv-2309-17453"},{"id":"swe-bench--swebench-com","name":"SWE-bench","url":"https://www.swebench.com/","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Benchmark","license":"MIT","latest_release":"last commit 2026-09-02","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/SWE-bench/SWE-bench (6117★)","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Benchmark of real GitHub issues that coding agents must resolve with a patch, with official leaderboards for Verified (500 human-filtered instances), Lite, Multimodal and Multilingual.","cited_cells":13,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/swe-bench--swebench-com"},{"id":"synapse--arxiv-2601-02744","name":"SYNAPSE","url":"https://arxiv.org/abs/2601.02744","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Spreading-activation graph + episodic-semantic split","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"graph","retrieval":"graph-traversal","persistence":"long-term","update":"extraction","unit":"episode","governance":"n/a","conflict":"llm-arbitrate","description":"Memory is a dynamic graph; retrieval fires spreading activation across edges rather than vector similarity. Lateral inhibition suppresses competing activations; temporal decay modulates edge weights. Episodic and semantic stores maintained separately.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/synapse--arxiv-2601-02744"},{"id":"terminal-bench--tbench-ai","name":"Terminal-Bench","url":"https://www.tbench.ai/","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-29","type":"Benchmark","license":"Apache-2.0","latest_release":"last commit 2026-07-11","deployment":"searched not found","pricing":"searched not found","mcp_support":"searched not found","a2a_support":"searched not found","backend_storage":"searched not found","llm_lock":"searched not found","github":"https://github.com/harbor-framework/terminal-bench-1 (2613★)","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Benchmark of agents completing hard tasks in a terminal environment, now at version 4.0, with a public leaderboard of model and agent pairs. It is hosted by Stanford, Harbor and the Laude Institute.","cited_cells":12,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/terminal-bench--tbench-ai"},{"id":"test-time-training-ttt--arxiv-2407-04620","name":"Test-time training (TTT)","url":"https://arxiv.org/abs/2407.04620","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Hidden states that update at test time","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"parametric","retrieval":"parametric-recall","persistence":"session","update":"parametric-edit","unit":"weight","governance":"opaque","conflict":"out-of-scope","description":"Stanford's Test-Time Training layers — RNN-like hidden state is itself an inner ML model that updates at test time on the current input.","cited_cells":5,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/test-time-training-ttt--arxiv-2407-04620"},{"id":"textworld--gh-microsoft-textworld","name":"TextWorld","url":"https://github.com/microsoft/TextWorld","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Text-adventure RL benchmark — procedurally-generated IF games","license":"MIT","latest_release":"","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"1.3k★ Python","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"Microsoft Research's TextWorld — a sandbox for training and evaluating RL agents on text-based games. Procedurally generates interactive-fiction games of tunable difficulty (map size, quest length, ingredients) with full ground-truth knowledge graphs. Used as the basis for the FirstTextWorld and TextWorld Challenge competitions.","cited_cells":30,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/textworld--gh-microsoft-textworld"},{"id":"tiermem--arxiv-2602-17913","name":"TierMem","url":"https://arxiv.org/abs/2602.17913","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Provenance-linked summary→raw escalation","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity|extraction-pull","persistence":"long-term","update":"append-only","unit":"summary","governance":"auditable","conflict":"n/a","description":"Two-tier: fast compressed summary index (Tier 1) + immutable raw-log store (Tier 2), with explicit provenance pointers linking summary units to source pages. Runtime sufficiency router escalates to raw log only when summary inadequate.","cited_cells":18,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/tiermem--arxiv-2602-17913"},{"id":"timem--arxiv-2601-02845","name":"TiMem","url":"https://arxiv.org/abs/2601.02845","section":"Research / specialised systems","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Temporal Memory Tree (5-layer)","license":"Custom","latest_release":"no releases","deployment":"searched not found","pricing":"searched not found","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"123★ Python","storage":"hybrid","retrieval":"similarity|hybrid-rerank","persistence":"long-term","update":"consolidation","unit":"episode","governance":"opaque","conflict":"llm-arbitrate","description":"Temporal-Hierarchical Memory Consolidation. 5-layer Temporal Memory Tree (segments → profiles); semantic-guided consolidation without fine-tuning; complexity-aware recall planning + gating. Jan 2026.","cited_cells":23,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/timem--arxiv-2601-02845"},{"id":"titans-google--research-google","name":"Titans (Google)","url":"https://research.google/blog/titans-miras-helping-ai-have-long-term-memory/","section":"Research / specialised systems","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Neural fast/slow + surprise gating","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"attention","persistence":"session","update":"parametric-edit","unit":"kv-token","governance":"opaque","conflict":"n/a","description":"Neural long-term memory module that learns to memorise at test time. Uses gradient-of-loss as \"surprise\" signal; adaptive weight-decay forgetting. Three variants: MAC (memory-as-context), MAG (memory-as-gate), MAL (memory-as-layer).","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/titans-google--research-google"},{"id":"toolformer--arxiv-2302-04761","name":"Toolformer","url":"https://arxiv.org/abs/2302.04761","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"research-code","last_verified_at":"2026-10-23","type":"Self-taught tool use","license":"MIT","latest_release":"0.0.30 2023-12-21","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"2.1k★ Python","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"skill","governance":"n/a","conflict":"llm-arbitrate","description":"Language models teach themselves to use tools.","cited_cells":26,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/toolformer--arxiv-2302-04761"},{"id":"toolgen--arxiv-2410-03439","name":"ToolGen","url":"https://arxiv.org/abs/2410.03439","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Tool retrieval + calling via generation","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"skill","governance":"n/a","conflict":"n/a","description":"Unified tool retrieval and calling via generation.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/toolgen--arxiv-2410-03439"},{"id":"toolllm--arxiv-2307-16789","name":"ToolLLM","url":"https://arxiv.org/abs/2307.16789","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"16k+ real-world API mastery","license":"Apache-2.0","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"5.6k★ +49/mo Python","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"read-only","unit":"skill","governance":"n/a","conflict":"llm-arbitrate","description":"Facilitates LLMs mastering 16,000+ real-world APIs.","cited_cells":24,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/toolllm--arxiv-2307-16789"},{"id":"toolmem--arxiv-2510-06664","name":"ToolMem","url":"https://arxiv.org/abs/2510.06664","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Learnable tool-capability memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"extraction","unit":"skill","governance":"n/a","conflict":"llm-arbitrate","description":"Enhances multimodal agents with learnable tool capability memory.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/toolmem--arxiv-2510-06664"},{"id":"transformer--arxiv-2501-06252","name":"Transformer²","url":"https://arxiv.org/abs/2501.06252","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"SVD + dynamic expert blending","license":"Apache-2.0","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"1.2k★ Python","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"skill","governance":"n/a","conflict":"llm-arbitrate","description":"Encodes procedural expertise using SVD decomposition with dynamic expert-vector blending.","cited_cells":25,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/transformer--arxiv-2501-06252"},{"id":"transformer-xl--arxiv-1901-02860","name":"Transformer-XL","url":"https://arxiv.org/abs/1901.02860","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"research-code","last_verified_at":"2026-10-23","type":"Segment-level recurrence + cached states","license":"Apache-2.0","latest_release":"no-release no-release","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"3.7k★ +60/mo Python","storage":"kv-cache","retrieval":"attention|cache-lookup","persistence":"session","update":"evict-oldest","unit":"kv-token","governance":"inspectable","conflict":"n/a","description":"Extends context through segment-level recurrence + caching of hidden states from prior segments. Foundational long-context architecture.","cited_cells":28,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/transformer-xl--arxiv-1901-02860"},{"id":"transformerfam--arxiv-2404-09173","name":"TransformerFAM","url":"https://arxiv.org/abs/2404.09173","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Feedback Attention Memory","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv-cache","retrieval":"attention","persistence":"session","update":"overwrite","unit":"kv-token","governance":"n/a","conflict":"n/a","description":"Google. Feedback attention loop enables the model to attend to its own latent representations — emerges as working memory. No extra weights; integrates with pre-trained models.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/transformerfam--arxiv-2404-09173"},{"id":"ufo2--arxiv-2504-14603","name":"UFO2","url":"https://arxiv.org/abs/2504.14603","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Desktop AgentOS","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity|agentic","persistence":"cross-session","update":"extraction","unit":"skill","governance":"user-controllable","conflict":"llm-arbitrate","description":"Desktop AgentOS with memory-driven task automation.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/ufo2--arxiv-2504-14603"},{"id":"v-jepa-2--arxiv-2506-09985","name":"V-JEPA 2","url":"https://arxiv.org/abs/2506.09985","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Video-JEPA + understanding/prediction/planning","license":"CC-BY-NC-4.0","latest_release":"no releases","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"3.8k★ Python","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"read-only","unit":"trajectory","governance":"n/a","conflict":"n/a","description":"Meta. Self-supervised video model. Predicts in latent space (no pixel decoder); supports understanding, prediction, and planning downstream. June 2025.","cited_cells":22,"third_party_cells":2,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/v-jepa-2--arxiv-2506-09985"},{"id":"v-jepa-original--arxiv-2404-08471","name":"V-JEPA (original)","url":"https://arxiv.org/abs/2404.08471","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Non-generative video world model","license":"CC-BY-NC-4.0","latest_release":"no releases","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"3.8k★ Python","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"read-only","unit":"trajectory","governance":"n/a","conflict":"n/a","description":"Meta AI. Predicts masked spatio-temporal video patches in latent representation space rather than pixel space. Implicit world model without generative decoding. Released for self-supervised video representation learning from VideoMix2M.","cited_cells":22,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/v-jepa-original--arxiv-2404-08471"},{"id":"voyager--arxiv-2305-16291","name":"Voyager","url":"https://arxiv.org/abs/2305.16291","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Lifelong-learning Minecraft agent","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"vector","retrieval":"similarity","persistence":"long-term","update":"accretion","unit":"procedure","governance":"plaintext","conflict":"n/a","description":"NVIDIA lifelong-learning agent for Minecraft — automatic curriculum + skill library + iterative prompting. Foundational procedural-memory paper.","cited_cells":6,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/voyager--arxiv-2305-16291"},{"id":"webarena--webarena-dev","name":"WebArena","url":"https://webarena.dev/","section":"Memory benchmarks & evaluation","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Web-agent benchmark — 812 tasks across 6 self-hosted sites","license":"Apache-2.0","latest_release":"","deployment":"not applicable — run locally","pricing":"not applicable — open benchmark; free to use","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"1.2k★ Python","storage":"n/a","retrieval":"n/a","persistence":"n/a","update":"read-only","unit":"n/a","governance":"n/a","conflict":"n/a","description":"CMU's WebArena — 812 hand-crafted web tasks across 6 self-hosted reproductions of real websites (Reddit-clone Postmill, GitLab, OneStopShop, CMS, Map, Wikipedia). Released ICLR 2024. Distinct from Mind2Web (real-website screenshots) — WebArena is fully reproducible with Docker-deployable backends. Spawned VisualWebArena, WorkArena, and dozens of agent benchmarks.","cited_cells":28,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/webarena--webarena-dev"},{"id":"webweaver--arxiv-2509-13312","name":"WebWeaver","url":"https://arxiv.org/abs/2509.13312","section":"Recent method papers — theorized, no distinct product","tier":4,"status":"active","last_verified_at":"2026-10-28","type":"Web-scale evidence + dynamic outlines","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"hybrid","retrieval":"similarity|agentic","persistence":"session","update":"overwrite","unit":"document","governance":"n/a","conflict":"append","description":"Structures web-scale evidence with dynamic outlines for research agents.","cited_cells":17,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/webweaver--arxiv-2509-13312"},{"id":"wise","name":"WISE","url":"","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Knowledge memory for lifelong editing","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"parametric","retrieval":"parametric-recall","persistence":"parametric-permanent","update":"parametric-edit","unit":"fact","governance":"inspectable","conflict":"llm-arbitrate","description":"Rethinks knowledge memory for lifelong model editing. NeurIPS 2024.","cited_cells":19,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/wise"},{"id":"workmate--journals-plos-org","name":"WorkMATe","url":"https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0316453","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-28","type":"Bio-inspired RL-gated memory slots","license":"","latest_release":"","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"","storage":"kv-cache","retrieval":"attention","persistence":"session","update":"agent-controlled","unit":"kv-token","governance":"n/a","conflict":"n/a","description":"Kruijne et al. RL-learned gating policies open/close multiple working-memory slots independently. Handles hierarchical 12-AX task with multiple concurrent context levels; all-or-nothing slot updates.","cited_cells":18,"third_party_cells":1,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/workmate--journals-plos-org"},{"id":"yarn--arxiv-2309-00071","name":"YaRN","url":"https://arxiv.org/abs/2309.00071","section":"Recent method papers — theorized, no distinct product","tier":3,"status":"active","last_verified_at":"2026-10-23","type":"Yet another RoPE extensioN","license":"searched not found","latest_release":"searched not found","deployment":"","pricing":"","mcp_support":"","a2a_support":"","backend_storage":"","llm_lock":"","github":"searched not found","storage":"kv-cache","retrieval":"attention","persistence":"session","update":"append-only|parametric-edit","unit":"kv-token","governance":"opaque","conflict":"out-of-scope","description":"RoPE-extension method for context-length scaling — combines NTK-aware interpolation with attention scaling, requires only short fine-tuning. Widely adopted in OSS long-context models.","cited_cells":6,"third_party_cells":0,"page":"https://mrpeppersdev.github.io/agent-infrastructure-landscape/systems/yarn--arxiv-2309-00071"}]}