Key Info
Tencent's Weixin AI team has developed WeAgent-MMSearch, a search agent that maintains persistent multimodal state so retrieved images remain accessible throughout the search trajectory, enabling cross-turn visual reasoning and reverse-image searches.
Highlights
- Persistent multimodal state keeps retrieved images accessible across the entire search process
- Agent can revisit images, cross-check visual evidence across turns, and perform reverse-image searches
- Designed to prevent multimodal search trajectories from collapsing back into text-only reasoning