Researchers from Sun Yat-sen University and China Mobile have introduced EdgeAgent, a tool designed for local multi-agent inference. According to the announcement, this development was made on 2026-10-02 within China. The company says the system allows multiple agents to operate locally without relying on distant cloud servers. Reported

What happened, in plain terms. It is filed as a research/economics story (region: China). The original source is arxiv.org: https://arxiv.org/html/2610.03394v1 Reported

The researchers note that the system tackles issues like shared-memory bandwidth contention and tool-induced stalls, noting a 1.77× speedup over batched speculative decoding under extreme tool-use latencies on Apple M4. Reported

The setup highlights that managing local agents depends on hardware-aware scheduling, though the outcome does not confirm power savings or compatibility with 27B-class Hermes models. Reported

It sits alongside other moves in the same feed, where UC Irvine-led researchers showed schedule-aware robot navigation, University of Iowa teams introduced lower-cost persistent memory, and Hubei expanded subsidies for various robot types. Reported

Whether the follow-up reports show real evidence on multi-agent coordination, tool orchestration and cloud-independent operation under constrained power and RAM, not only launch claims. Orion's reading

Runway has shared news of its Praxis-1 robot-policy model and early hardware-partner testing, noting the details on runway.com as of 2026-09-30. Reported

California has introduced new requirements for autonomous-vehicle incident response and remote operations, details confirmed by the state legislature on 2026-09-30 via leginfo.legislature.ca.gov. Reported

Meanwhile, researchers from the Université de Montréal and Mila have noted that faster reasoning can be achieved through selective latent-memory recall, a finding shared on 2026-09-30 via arxiv.org. Reported