AWS proposes MixUA: multi-granularity memory scoping for multi-user multi-agent systems
AWS · hf · 2026-11-14
AWS Agentic AI and USC present MixUA, a memory system that distills lessons from independent user-agent trajectories and assigns each to the right scope — global (shareable across all users and agents), user (private preferences following the user), or agent-specific — balancing cross-user/cross-agent learning with privacy, evaluated on the new CrossLearnBench benchmark.
More from coding & agent
- Gemini 4 Hydrogen goes GA with agentic coding upgrades, lands in GitHub Copilot — feblet · 2026-11-14
- Unpopular opinion on Reddit: most AI agents are just overcomplicated chatbots — Charming_Cod_2910 · 2026-11-14
- Dev demos real computer use: agent drives your actual Mac, voice-first — RileyRalmuto · 2026-11-14
- One-line prompt + image tool: Fable 5.2 and Seedream 5.5 ship a real-art browser game for ~$3 — Fun_Walk_4965 · 2026-11-14
- Prefix-reuse FLOPs: a new metric to measure an agent's true compute cost across context edits — RulinShao · 2026-11-14
- Viral 'caveman speak' token-saver launches Caveman Code, claims 25-50% off coding bills — ycombinator · 2026-11-14