entity Updated 2026-07-07 Tags: Ai-Tool, Agent, Startup

AirJelly

AirJelly is the proactive context-aware personal-agent product discussed in OpenClaw 之后,谁将定义主动式 AI 的新战场?|对谈 AirJelly 黄柏特. Founder Huang Bote describes it as a partner that remembers what the user is doing across places, understands task state, and appears at the right moment to help rather than waiting for a fully specified prompt.

The product descends from Mycontext, but changes the capture strategy from periodic screenshots toward event- and intent-triggered context. Its signature mechanism is treating Enter as an Intent Context signal across IM, chatbot, and search workflows; when the user submits something, AirJelly captures the surrounding screen and then organizes the result into events and entities for Persistent Agent Memory.

Key Points

  • AirJelly combines OS-Level Context, memory, and execution ability: it can read documents and code, call skills, use a browser, operate a computer, and use stored context.
  • Open Claw and Claude Code influenced the product, but also pushed the team away from simplified task-execution tooling toward context acquisition, storage, and recall as the harder layer.
  • The product tries to make Proactive Agents task-continuing rather than distracting: suggestions should extend the user’s current intent instead of adding loosely related information.
  • Privacy is a central risk because the product touches screenshots and personal context; the source says AirJelly plans local storage for images/virtual context, end-to-end encryption, and PII desensitization.
  • The team is also experimenting with an internal team version where personal agents coordinate in a group, while keeping sharing voluntary rather than turning AirJelly into monitoring software.

Connections