ANTMAN: Adaptive Need Tracking for Multi-Agent Navigation in Large Information Spaces
ANTMAN coordinates agents around unresolved needs, not fixed chunks of a search space.
The paper describes a revisable “Need Graph” that tracks requirements, evidence, failed attempts, and progress as agents search. That graph drives worker selection, routing, and local recovery when new evidence changes what remains to be found. In tests covering multi-document QA, long-context scaling, and structured navigation, the framework still worked when smaller models handled delegated tasks. With searchable context increased 16x, ANTMAN’s active coordination rose 1.23x, versus more than 15x for partition-based baselines. HF Daily Papers' note
The paper describes a revisable “Need Graph” that tracks requirements, evidence, failed attempts, and progress as agents search. That graph drives worker selection, routing, and local recovery when new evidence changes what remains to be found. In tests covering multi-document QA, long-context scaling, and structured navigation, the framework still worked when smaller models handled delegated tasks. With searchable context increased 16x, ANTMAN’s active coordination rose 1.23x, versus more than 15x for partition-based baselines. HF Daily Papers' note
score 4