Research / specialised systems

10 systems in the research / specialised systems category of the AI Agent Infrastructure Landscape, grouped by maturity tier.

Tier 2 — production-ready (1)

  • MemPalace Verbatim chunks + spatial metaphor

    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.

Tier 3 — emerging (5)

  • A-MEM Atomic-note / Zettelkasten-style

    Treats memories as atomic linkable notes — explicit nod to Zettelkasten knowledge management. Dynamic linking; retroactive memory revision.

  • BAI-LAB MemoryOS Hierarchical: short / mid / long-term

    Hierarchical "OS" with Storage / Updating / Retrieval / Generation modules. Short-term → mid-term via FIFO dialogue-chain; mid-term → long-term via segmented paging.

  • EVOLVE-MEM Self-adaptive 3-level hierarchy

    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).

  • MemoRAG Global memory-enhanced RAG

    RAG framework on top of a long-context memory model. Builds global memory once, generates contextual clues at query time. TheWebConf 2025.

  • MemOS (MemTensor) "Memory operating system" + MemCubes

    Treats memory as an OS-managed resource with explicit allocation and process-like scoping. MemCubes unify parametric, activation, and plaintext memory.

Tier 4 — early / experimental (4)

  • EverMemOS Episodic + semantic + reconstructive

    Self-organizing memory OS for structured long-horizon reasoning. Three-phase model: episodic, semantic, reconstructive.

  • LiCoMemory CogniGraph + temporal/hierarchy-aware retrieval

    Lightweight hierarchical graph (CogniGraph) with entities and relations as semantic indexing layers. Incremental graph construction, fast updates, low-latency inference. Nov 2025.

  • TiMem Temporal Memory Tree (5-layer)

    Temporal-Hierarchical Memory Consolidation. 5-layer Temporal Memory Tree (segments → profiles); semantic-guided consolidation without fine-tuning; complexity-aware recall planning + gating. Jan 2026.

  • Titans (Google) Neural fast/slow + surprise gating

    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).