StreamVibe brings proactive video agents to evolving user intent
The StreamVibe paper introduces StreamVibe-Bench, a benchmark with 4.1K evaluation samples and 85K training samples covering explicit, implicit, and dynamic intents. It also presents StreamVibe-VL, a streaming vision-language model that infers implicit intent from continuous observations, and StreamVibe-Harness, a hierarchical-memory framework for tracking changing user and environmental states. The authors report a 10.7% improvement in answer quality and response timing over baselines, plus a 12.3% gain in proactive interaction from the harness.
The work directly addresses how vision-language agents can respond to unstated user intent and adapt during sustained interactions, although its reported benchmark gains come from the paper’s authors.
Computer Science > Computer Vision and Pattern Recognition
arXiv:2610.34662 (cs)
[Submitted on 26 Oct 2026]
Title:StreamVibe: Proactive Video Agents for Evolving User Intent
Authors:Chengzhi Liu, Yuxiang Lai, Yuzhe Yang, Yanqing Liu, Jiheng Hou, Yue Cao, Haonan Wang, Ruixi Fan, Lan Wang, Sheng Liu, Songyou Peng, Yale Song, Kevin Qinghong Lin, Xin Eric Wang, Jui-Hsin Lai, Youbao Tang
View a PDF of the paper titled StreamVibe: Proactive Video Agents for Evolving User Intent, by Chengzhi Liu and 15 other authors
#StreamVibe#StreamVibe-Bench#StreamVibe-VL#StreamVibe-Harness