MATM is a framework for population-level reuse of agent trajectories. Producer agents contribute trajectories from their task executions to a shared repository, and consumer agents retrieve those trajectories to improve their own task execution. The framework is tested in interactive environments, ALFWorld and WebArena, where trajectories are long and encode especially rich procedural structure.
Retrieving trajectories from MATM improved downstream task performance and reduced interaction steps, without requiring coordination among agents or joint training. This matters because agent trajectories are usually discarded after one use or kept private to the agent that produced them, so newly instantiated agents repeatedly rediscover solutions that already exist elsewhere in the ecosystem. The result suggests a simple infrastructure pattern for experience sharing across heterogeneous, independently deployed agents.