hyperstack

Cloud memory for AI agents. Store knowledge as small cards (~350 tokens) instead of stuffing conversation history (~6,000 tokens) into every prompt. 94% token savings. Hybrid semantic + keyword search. Zero LLM cost for memory ops.