SearchOS-V1: Towards Robust Open-Domain Information-Seeking Agent Collaboration
SearchOS is a system-level multi-agent framework designed to address the problem in open-domain information retrieval where agents struggle to track task progress due to growing interaction history and fall into repetitive loops. The framework transforms implicit search progress into explicit, persistent, shared states, including relational schema completion, search-oriented context management (SOCM), and pipeline parallel scheduling.
SearchOS externalizes search state into frontier tasks, evidence graph, coverage graph, and failure memory via SOCM, and employs pipeline parallel scheduling to continuously fill idle slots to address uncovered search gaps.
The core innovation of SearchOS is explicitly modeling search progress as a shareable state and avoiding repetitive loops in single-agent or simple multi-agent systems through pipeline parallel scheduling. A signal to verify next is whether SOCM can maintain search efficiency without degradation in longer interaction histories (e.g., 100+ turns) or more complex tasks (e.g., multi-table join queries).
This work indicates that information retrieval agents are evolving from single-agent tool invocation to system-level multi-agent collaboration, with explicit state management becoming a key bottleneck.
SearchOS can reduce search costs (by reducing redundant calls) and improve final output quality for information retrieval agents, offering direct commercial value for knowledge-intensive industries such as legal, medical, and scientific research automation.
More frameworks that externalize search state may emerge in the future, potentially giving rise to middleware or platforms specialized for search state management.