← Back to research

Research Brief · 2026-11-10 Core

Deep Personalization: Continuous Context Observation for Local Proactive Agents

The idea, in brief

A local desktop agent learns from thumbs, dismissals and missed moments when this particular person wants help. Simulated-persona benchmark plus a week with 10 users.

Editorial summary · not the author abstract. Read the linked paper for the complete abstract and claims.
Figure 1: an Observer Agent watches desktop context, decides when to trigger an execution agent, and adapts on device from engagement, dismissal and missed opportunities.
Figure 1: an Observer Agent watches desktop context, decides when to trigger an execution agent, and adapts on device from engagement, dismissal and missed opportunities. Original figure · paper authors ↗

Release evidence

The Coco desktop app is public under Apache-2.0. The paper points the adaptation pipeline and OSWorld-Persona to an NVlabs repository that returned 404 on November 15; no adapters are released and no user-activity data will be.

Last checked · 2026-11-15