MIRIX-2.0: Multi-Agent Memory System
Multi-agent take on long-term agent memory, avoiding lossy consolidation and an ever-growing record store.
6 items · 2 sources · 5 days
Operational story trace
Follow in this browser to see new updates on your Live feed.
Latest change
The latest paper reports a multi-agent system deployed in production to keep customer-facing documentation in sync with fast-moving code.
This is a pattern across six independent arXiv papers, not one event: separate teams keep splitting agent work across cooperating roles. It opened with a multi-agent memory design and a crypto-trading system, then a safety paper argued Codex- and Claude Code-style multi-agent setups fall to attacks single frontier models resist.
Arc
Multi-agent take on long-term agent memory, avoiding lossy consolidation and an ever-growing record store.
Domain case: reflective agents plus market feedback for non-stationary crypto trading.
The security counterpoint: a break that exists only at the multi-agent level, shown on Codex- and Claude Code-style systems.
Pushes the pattern to automating the full reinforcement-learning training pipeline.
Applies feedback-scheduled agents to turning protocol descriptions into Tamarin formal models.
First item in the thread described as deployed: agents that keep customer docs current with code.
Multi-agent take on long-term agent memory, avoiding lossy consolidation and an ever-growing record store.
Domain case: reflective agents plus market feedback for non-stationary crypto trading.
The security counterpoint: a break that exists only at the multi-agent level, shown on Codex- and Claude Code-style systems.
Pushes the pattern to automating the full reinforcement-learning training pipeline.
Applies feedback-scheduled agents to turning protocol descriptions into Tamarin formal models.
First item in the thread described as deployed: agents that keep customer docs current with code.
What to watch — open questions
Storylines are threaded mechanically from the feed: stories that share a distinctive anchor across multiple days and sources. Each item links to its original source. The evidence trace, current state, and open questions are written by the editor routine and refreshed whenever a new beat lands.