Research project

CurveMemory is a promising routing experiment.

CurveMemory explores geometric, CPU-first routing for AI memory and large knowledge spaces. It is not the current product headline. The commercial focus is Trajbl: a post-retrieval context compressor that can plug into existing RAG and agent stacks.

0.109 msCPU routing latency
0.02 MBRouting RAM delta
55.5 KBCompact routing index
0GPUs required for routing
Research status

Useful ideas, not the main market claim.

CurveMemory may later contribute routing, ingest, or memory-classification ideas. For now it is presented as a research track behind the primary Trajbl product.

Geometric routing

Explores routing a query into compact semantic regions before broader retrieval work begins.

CPU-first design

Early measurements show very small routing latency and memory footprint.

Ingest-aware thinking

The project contains ideas around organizing memory before query time.

Research track

Promising, but not yet the product claim we want to put in front of customers.