Tag
10 posts tagged llm.
The advertised context window is 2 to 8 times larger than the effective context for multi-hop work, and 50 to 100 times larger than the effective context for reasoning. The number on the slide is the size of the door. The number that does work is the size of the room.
Nineteen agent-memory systems quietly reversed their biggest design choice: they stopped injecting context and gave the agent tools instead. Here is why.
Most agents are not running out of tokens. They are wasting them. Six mechanisms from 19 systems for keeping context budgets under control.
A flat memory store is the wrong default. How 7 of 19 agent-memory systems converged on tiered storage, what the heat formula looks like, and the 7 steps to get there.
Hybrid retrieval with RRF at k=60 is the consensus across 19 agent-memory systems. Here is what the reference implementation looks like and why the outlier matters.
A fact without provenance is an island. Six levels of agent-memory provenance across 19 systems - identity, source, confidence, versioning, causal, reciprocal.
The highest-ROI pattern across 19 agent-memory systems. Pay at write time, read for free. Six forms, why they compose, and the order that pays back fastest.
Six deployment shapes for agent memory. The architecture everyone debates isn't what decides how it feels to use. An honest look at Supermemory's bet.
19 agent-memory systems resolve into 8 architectural paradigms, not 8 flavours of one recipe. Eight different bets on what memory is. Pick the one that fits.
Nineteen open-source agent-memory systems agree on one thing, and it's what not to do. Vector RAG isn't enough on its own. Here's what they all added.
