Intent Context

Intent context is the high-signal slice of user context captured when a user is explicitly trying to do something. In OpenClaw 之后，谁将定义主动式 AI 的新战场？｜对谈 AirJelly 黄柏特, AirJelly uses Enter as the practical trigger: when the user submits a message, chatbot prompt, or search query, the surrounding screen and task state are more likely to reveal current intent than a random periodic screenshot.

The concept sits between OS-Level Context and Persistent Agent Memory. OS-level capture supplies raw evidence, but intent context tries to identify the moments worth preserving as events, entities, and task progress. The episode argues that this is what lets Proactive Agents help along the current task’s extension rather than create generic reminders or distracting suggestions.

Key Claims

• More context is not automatically better; indiscriminate recording can bury useful signals in noise.
• Intent moments are valuable because the user has just expressed a goal, constraint, or next action.
• Proactive help should be grounded in current intent and context, not only in a profile, schedule, or broad curiosity model.
• Intent context must still be bounded by privacy, permission, local storage, encryption, and user control.
• Durable memory depends on converting intent context into maintained events and entities rather than storing every screenshot as equal-value history.

Connections

• AirJelly and Huang Bote — product and founder that make the concept concrete.
• OS-Level Context — raw environment context from which intent context is selected.
• Persistent Agent Memory — durable store that intent context feeds.
• Proactive Agents — agent behavior that needs intent context to avoid becoming interruption.
• Context Engineering and Human-Agent Collaboration — broader practices for making agent assistance useful.