Mustafa Arslan
@mustafarslan.bsky.social
Pinned
Aeon: High-Performance Neuro-Symbolic Memory Management for Long-Horizon LLM Agents
Large Language Models (LLMs) are fundamentally constrained by the quadratic computational cost of self-attention and the "Lost in the Middle" phenomenon, where reasoning capabilities degrade as contex...
arxiv.org
New paper on arXiv: Aeon - High-Performance Neuro-Symbolic Memory for LLM Agents
Fixes "Vector Haze" with:
• Memory Palace spatial index
• SLB cache (<1ms latency)
• Episodic graph for long-horizon recall
C++23 core + Python shell for real autonomy
📄 Paper : arxiv.org/abs/2601.15311
Fixes "Vector Haze" with:
• Memory Palace spatial index
• SLB cache (<1ms latency)
• Episodic graph for long-horizon recall
C++23 core + Python shell for real autonomy
📄 Paper : arxiv.org/abs/2601.15311
I love seeing how the community visualizes the concepts from my Aeon paper! 🎨
Someone recently created these incredible slides breaking down my ‘Vector Haze’ concept and the ‘Cognitive OS’ architecture.
While I didn’t commission these, they are too good not to share
🧵 1/5
Someone recently created these incredible slides breaking down my ‘Vector Haze’ concept and the ‘Cognitive OS’ architecture.
While I didn’t commission these, they are too good not to share
🧵 1/5
January 29, 2026 at 6:12 AM
I love seeing how the community visualizes the concepts from my Aeon paper! 🎨
Someone recently created these incredible slides breaking down my ‘Vector Haze’ concept and the ‘Cognitive OS’ architecture.
While I didn’t commission these, they are too good not to share
🧵 1/5
Someone recently created these incredible slides breaking down my ‘Vector Haze’ concept and the ‘Cognitive OS’ architecture.
While I didn’t commission these, they are too good not to share
🧵 1/5
New paper on arXiv: Aeon - High-Performance Neuro-Symbolic Memory for LLM Agents
Fixes "Vector Haze" with:
• Memory Palace spatial index
• SLB cache (<1ms latency)
• Episodic graph for long-horizon recall
C++23 core + Python shell for real autonomy
📄 Paper : arxiv.org/abs/2601.15311
Fixes "Vector Haze" with:
• Memory Palace spatial index
• SLB cache (<1ms latency)
• Episodic graph for long-horizon recall
C++23 core + Python shell for real autonomy
📄 Paper : arxiv.org/abs/2601.15311
Aeon: High-Performance Neuro-Symbolic Memory Management for Long-Horizon LLM Agents
Large Language Models (LLMs) are fundamentally constrained by the quadratic computational cost of self-attention and the "Lost in the Middle" phenomenon, where reasoning capabilities degrade as contex...
arxiv.org
January 28, 2026 at 1:50 PM
New paper on arXiv: Aeon - High-Performance Neuro-Symbolic Memory for LLM Agents
Fixes "Vector Haze" with:
• Memory Palace spatial index
• SLB cache (<1ms latency)
• Episodic graph for long-horizon recall
C++23 core + Python shell for real autonomy
📄 Paper : arxiv.org/abs/2601.15311
Fixes "Vector Haze" with:
• Memory Palace spatial index
• SLB cache (<1ms latency)
• Episodic graph for long-horizon recall
C++23 core + Python shell for real autonomy
📄 Paper : arxiv.org/abs/2601.15311