Fetching from the wire…
01
02
03
04
05
06
07
08
Research2026-08-14 · source-backed
arXiv 2608.11879 benchmarked Mem0, Hindsight and Mastra Observational Memory across conversations up to 400 turns and 665 LoCoMo questions. Cost models built on conversation length miss badly because internal memory behavior dominates. Break-even against just replaying the full transcript varies wildly: some systems win within tens of turns, others never do inside 400. Accuracy spans 21-54%. Measure break-even against naive replay on your own traffic before adopting a memory layer.
Each link below shares sources, entities, or timing with this story.
Memori turns agent execution traces into structured persistent state, outperforming Zep, LangMem, and Mem0 on the LoCoMo benchmark while reducing prompt size by 67% vs Zep. Python and TypeScript SDKs. If you're building agents that need memory, benchmark this against whatever...
arXiv 2607.27090 targets the hidden tax in Mem0, MemGPT, and Zep: retrieved memory facts get re-prefilled on every single request. It precomputes each fact's KV representation, stores it on-GPU with a semantic embedding, and injects it into vLLM's paged cache via the KV-connec...
arXiv 2607.29032 converts sparse historical hidden states into reusable memory representations via a gating network intervening on current hidden states, all at inference time with no context reprocessing. Training uses evidence-conditioned self-distillation: a memory-augmente...
MemPalace (57,821 stars, v3.6.0) reports 96.6% raw recall@5 on LongMemEval with no LLM required, 98.4% with hybrid v4 on a held-out 450 questions, LoCoMo R@10 rising 60.3% → 88.9%, ConvoMem 92.9%, MemBench 80.3%, while explicitly refusing head-to-head comparison against Mem0,...
arXiv 2608.03463 sorts dialogue by compressibility, temporal dynamics, and fidelity requirement, storing informative segments as compact profile memory, temporally structured event memory, or source-grounded record memory, then updating only the evolving event memories during...
Gaurav Dadhich reframes agent memory from storage-and-retrieval into a lifecycle discipline with five primitives — architecting, ingesting, scoping, anticipating, and compacting & consolidation — across an organizational scope hierarchy rather than a single user. The economic...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.