Views
No views yet
Position IS Relationship - A Spatial Memory Fabric for AI Systems
Traditional: State → Project → Index → Retrieve → Reconstruct
(lossy at each step)
ARMS: State → Store AT coordinates → Retrieve → Inject directly
(native representation preserved)
| Primitive | Type | Purpose |
|---|---|---|
| Point | Vec<f32> | Any dimensionality |
| Proximity | fn(a, b) -> f32 | How related? |
| Merge | fn(points) -> point | Compose together |
| Place | fn(point, data) -> id | Exist in space |
| Near | fn(point, k) -> ids | What's related? |
1use arms_core::{Arms, ArmsConfig, Point};
2
3// Create ARMS with default config
4let mut arms = Arms::new(ArmsConfig::new(768));
5
6// Place a point in the space
7let point = Point::new(vec![0.1; 768]);
8let id = arms.place(point, b"my data".to_vec()).unwrap();
9
10// Find nearby points
11let query = Point::new(vec![0.1; 768]);
12let neighbors = arms.near(&query, 5).unwrap();
┌─────────────────────────────────────────────────────────────┐
│ ARMS │
├─────────────────────────────────────────────────────────────┤
│ CORE (pure math, no I/O) │
│ Point, Id, Blob, Proximity, Merge │
│ │
│ PORTS (trait contracts) │
│ Place, Near, Latency │
│ │
│ ADAPTERS (swappable implementations) │
│ Storage: Memory, NVMe (planned) │
│ Index: Flat, HAT (see arms-hat crate) │
│ │
│ ENGINE (orchestration) │
│ Arms - the main entry point │
└─────────────────────────────────────────────────────────────┘
| Hippocampus | ARMS |
|---|---|
| Encodes episodic memories | Stores attention states |
| Spatial navigation | High-dimensional proximity |
| Pattern completion | Near queries |
| Memory consolidation | Merge operations |
| Place cells | Points at coordinates |
arms-hat - Hierarchical Attention Tree index adapter (100% recall, 70x faster than HNSW)arms-nvme - Persistent storage via memory-mapped filesarms-distributed - Sharded storage across machinesarms-gpu - CUDA-accelerated similarity searcharms-py - Python bindings1[dependencies]
2arms-core = "0.1"paper/ directory.1@article{young2026arms,
2 author = {Young, Andrew},
3 title = {ARMS: A Spatial Memory Fabric for AI Systems},
4 journal = {arXiv preprint},
5 year = {2026},
6 url = {https://github.com/automate-capture/arms}
7}