Saccade

Live semantic Truth Layer for authorized Chrome and Edge tabs.

Saccade connects compatible AI agents to browser tabs you open for them or explicitly share. It gives the agent structured page controls and meaningful changes, so it can help fill forms, navigate supported controls and check results in your browser. Setup requires the Saccade Chrome extension and the local Node.js MCP package, @nanlogic/saccade, configured in a compatible AI client. Version 0.4.20 adds experimental browser-first semantics: browser-computed control roles, names and states take priority, with DOM information supplying geometry and other details. This uses Chrome's debugger permission only on authorized tabs. Chrome may show a debugging notice. If browser semantics are unavailable, Saccade reports the limitation instead of claiming equivalent coverage. Experimental video observation supplies sampled frames with measured timestamps and available captions. It does not watch every frame or transcribe audio. Brief events may be missed, and protected or unsupported media may not be readable. Experimental 3D observation supplies stills and short frame sequences for applications using the Saccade scene bridge, alongside application-reported motion state. It does not inspect arbitrary WebGL objects or guarantee correct animation diagnosis. Privacy and control: - Tabs are private until opened by an agent or explicitly shared with it. - One observation approval covers that agent's authorized tabs for its live session; other agents and unshared tabs are excluded. - Stop sharing revokes access to a tab. - Video observation may temporarily pause or seek playback. Screenshots and frames may contain personal information. - Diagnostics do not persist screenshots, media recordings, cookies or editable field contents. - Supported protected fields are withheld from semantic page reads. - Saccade does not expose DOM selectors, arbitrary scripts, debugger commands or arbitrary-coordinate actions to agents. Browser semantics, video and 3D features are experimental. The broader lifecycle test matrix and performance comparison remain incomplete. No full control coverage, benchmark-level accuracy, speed advantage or token savings are claimed. Developed and published by NaN Logic LLC.