Quantum Agent Memory

QAOA-powered memory management for AI agents. Three quantum layers replace classical heuristics for clustering, compaction, and recall.

Companion to Quantum Memory Graph (QMG) v1.2 — #1 on LongMemEval (98.6% R@5).

Three Quantum Layers

Layer 1: Clustering

Group N memories into coherent clusters via balanced graph-cut QAOA.

• Cost matrix: temporal (25%), relational (30%), categorical (25%), recency (20%)
• 100% optimal for n≤14, speed crossover at n=20

Layer 2: Compaction

Select optimal K memories to keep from M total.

• Maximizes coverage + coherence + value + recency with budget penalty
• Beats greedy selection by ~1% consistently

Layer 3: Recall

Find the best K memories for a query — optimizes for synergy, not just individual relevance.

• Finds memory combinations that Top-K similarity search misses
• Individual relevance (40%) + pairwise synergy (30%) + diversity (20%) + recency (10%)

Quick Start

git clone https://github.com/Dustin-a11y/quantum-agent-memory.git
cd quantum-agent-memory
python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt
python -m quantum_agent_memory benchmark

Integration with QMG v1.2

Point the benchmark at a live QMG instance:

python -m quantum_agent_memory benchmark --qmg-url http://localhost:8503

For OpenClaw agent integration, see references/openclaw-plugin.md.

IBM Quantum Hardware

Submit circuits to real IBM quantum processors:

pip install qiskit-ibm-runtime
python -m quantum_agent_memory submit --ibm-token YOUR_TOKEN

Benchmarking

Run the full 3-layer benchmark:

python -m quantum_agent_memory benchmark

Expected output:

• Clustering: ~98-100% optimal
• Compaction: 100% optimal
• Recall: 100% optimal, quantum finds synergistic combos Top-K misses
• Avg accuracy: ~99.7%

Requirements

• Python 3.10+
• Qiskit 2.0+
• qiskit-aer (simulation)
• numpy

Optional:

• qiskit-ibm-runtime (real hardware)
• sentence-transformers (embeddings)