Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-08-16 · high
It builds memory from recall cues instead of one-shot retrieval and lifts benchmark accuracy up to 11.87%.
Why now: The paper surfaced in the August 16, 2026 briefing, and no outside lab has weighed in on the numbers yet.
RippleMem cuts the cost of building an agent's long-term memory graph by about 30-fold. It builds that graph from recall cues instead of one-shot retrieval, per a preprint posted to arXiv.
That's the gap between a memory system cheap enough to run on every conversation and one stuck as a research demo. The same preprint reports LLM-as-Judge accuracy gains up to 11.87% on long-conversation benchmarks.
The bottleneck isn't storage, per the paper's argument. It's finding evidence scattered across many past interactions. Flat retrieval returns isolated records with no connections between them, and full graph memory systems cost too much to build for every conversation.
To do that, it stores conversation history as cue-rich episodic units inside an event-centric graph. It recalls anchor points through hybrid cues, then expands along semantic and structural links. Each recalled memory becomes a cue for finding more, per the paper.
Independent groups haven't reproduced the LongMemEval-S numbers yet.
Each link below shares sources, entities, or timing with this story.
Simon Willison released LLM / Shared entity: LLM / Shared topic / Earlier coverage
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; overlapping topics (agent, cost).
Simon Willison released LLM / Shared entity: LLM / Earlier coverage / Tension
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-07-27.
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-06-19.
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-06-18.
Simon Willison released LLM / Shared entity: LLM / Earlier coverage
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-06-19.
Simon Willison released LLM / Shared entity: LLM / Tension
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; pushes against this story (pushback).
Simon Willison released LLM / Same source domain / Shared topic
Linked by a graph relationship (Simon Willison released LLM); reported by the same outlet (arxiv.org); overlapping topics (agent, retrieval).
mempalace benchmarked against LongMemEval / Shared entities / Earlier coverage / Tension
Linked by a graph relationship (mempalace benchmarked against LongMemEval); both cover LLM, LongMemEval; earlier LLM coverage from 2026-07-28.