AgensFlow: A Coordination-Policy Substrate for Multi-Agent Systems

Researchers introduced AgensFlow, an open-source substrate for multi-agent coordination that keeps models frozen and learns routing, role, skill, and topology decisions.

Its coordination policy is built from inspectable state-action statistics, making multi-agent execution serializable and auditable.

In tests on distributed-systems incident response and security-advisory tasks, AgensFlow outperformed a fixed pipeline on coordination-heavy problems.

The framework aims to make multi-agent AI more flexible while preserving control and transparency.