Open-Source Proactive Assistant Coco: Building a Human-AI Collaboration Layer

EchoShao8899 introduced the open-source project Coco, positioning it as a proactive human-AI collaboration layer rather than just another powerful agent. Based on human research in real-world tasks, the author argues that collaboration outcomes heavily depend on the human's state and capabilities. Since many people don't know what AI can help with, the focus should shift towards "how to collaborate" rather than simply making models stronger.

Core Design

Coco is described as a proactive collaborative assistant running as a lightweight desktop app, designed to intervene at the right time and stay quiet when it shouldn't. The core component is the Computer Use Behavior Observation Protocol, a microservice used to capture, classify, and pass user computer usage behavior to downstream systems. The author highlights two main capabilities: Proactive Support, which offers reminders or suggests better tools when a user is stuck; and Smart Orchestration, which breaks down goals into subtasks and distributes execution between humans and agents.

Roadmap and Collaboration

The author states that Coco aims to make AI more accessible to people across different fields and skill levels. To reduce ongoing API costs, the team is working with hardware experts to make Coco "on-device native." Additionally, they are collaborating with the GUM project to improve preference management and memory. The overall direction is to evolve increasingly strong AI capabilities from mere software demos into a collaboration infrastructure closer to daily work.

2026-07-14 ~ 2026-07-14 · 6 related posts

1 near-duplicate retellings: EchoShao8899