Continuum Memory Architectures for Long-Horizon LLM Agents#

Original | Raw

Academic paper by Joe Logan (Mode7 GK, Tokyo) defining CMA as a formal architectural class for agent memory. Provides the theoretical foundation for why RAG is insufficient and what properties a real memory system must have.

The Problem with RAG#

RAG treats memory as a stateless lookup table: information persists indefinitely, retrieval is read-only, temporal continuity is absent. Agents reconstruct context afresh each time rather than maintaining it — no continuity of identity or purpose.

Six CMA Requirements (Necessary Conditions)#

  1. Persistence — state preserved across sessions. Agent accumulates identity.
  2. Selective Retention — memories compete for accessibility (recency, usage, salience). Ebbinghaus forgetting curves.
  3. Retrieval-Driven Mutation — every lookup alters future accessibility. Consulted fragments stabilize, contradictory ones recede.
  4. Associative Routing — structure connecting entities. Activation spreads along links for multi-hop answers.
  5. Temporal Continuity — episodic traces defined by order. Explicit temporal edges enable “what was happening around X?” queries.
  6. Consolidation and Abstraction — sleep-inspired replay and gist extraction. Episodes fade once schemas emerge.

Standard RAG meets none of these. These are necessary and collectively sufficient.

Reference Lifecycle#

Ingest → Retrieval → Mutation → Consolidation. Memory substrate is a structured store with semantic, temporal, and structural edges. Each node retains reinforcement history, salience, timestamps, provenance.

Evaluation Results (vs RAG Baseline)#

CMA won 82 of 92 decisive trials across four behavioral probes. Latency increased ~2.4× (1.48s vs 0.65s).

StudyRAG WinsCMA WinsEffect Size
Knowledge Updates138d=1.84
Temporal Association113h=2.06
Associative Recall514h=0.99
Disambiguation317h=1.55

Failure Modes#

  • Latency scaling with graph edges
  • Memory drift from reinforcement feedback loops
  • Temporal sensitivity (half of temporal queries stumped both systems)
  • Interpretability of evolving graphs
  • Data governance for persistent memories

Key Insight#

Future agent architectures differentiated less by model size, more by how they instantiate memory. Memory is an inevitable architectural primitive.

mem0 (Mem0g), MemGPT, A-MEM (Zettelkasten, NeurIPS 2025), Hindsight (four-network architecture), SimpleMem, MemoRAG.

See Also#