AirJelly launches a desktop proactive AI assistant, using the Enter key as an anchor to capture user intent

ME News message: On April 22 (UTC+8), according to Beating Monitoring, the desktop AI assistant AirJelly launched by Bert, the founder of Continuous Low Entropy (Low Entropy AI) and a former product manager at ByteDance, has secured funding from Source Capital.

Bert previously led the context engineering product MineContext at ByteDance. He then left with his original team to start a business. The team consists of 11 members, all working offline in Beijing.

AirJelly does not perform full-screen recording. Instead, it uses the Enter key as an anchor: each time the user presses Enter, it captures the screen, and—via Accessibility permissions—collects the current application, the input box type, and the context. It models this information as an Event, which is then summarized into a Task by AI.

Compared with the previous MineContext full-screen approach that captured about 1,500 screenshots per day, the Enter mechanism compresses the number of screenshots to about 300, reducing costs to one-fifth, while also lowering misjudgments caused by irrelevant screenshots.

The memory system has two layers: static information is modeled as Entities (people, projects, etc.), and dynamic information is modeled as Tasks. Each Task includes a title, summary, progress, next step, and related Events.

For recall, it combines vector search, keyword search, and time-decay weighting, with all data stored locally. On this basis, AirJelly determines whether a Task update has reached the push threshold, and proactively recommends the next action to users—a concept the team calls Proactive Trigger.

The underlying agent execution capability connects to OpenClaw’s Pi framework, together with MineContext’s screen understanding ability. The product currently supports macOS, and the Windows and Linux versions are under development. It is available for free download at airjelly.ai.

Next, the team plans to launch the “Next Enter Prediction” feature, which predicts what content the user will enter on their next Enter based on historical behavioral trajectories, and also introduce a team version to support collaboration among multiple people.
(Source: BlockBeats)