now

currently working on

visor v2 is the next direction for visor and is currently taking shape.

it explores a Model Context Protocol approach for browser agents, with the aim of helping them communicate with websites and interact with frontend interfaces through useful structured context.

the idea is to move past giving an agent an entire webpage and hoping it develops manners.

future

visor v2 should become a clearer bridge between an agent and a live interface.

the goal is for an agent to understand what it can safely touch, what it should leave alone, and what an action actually means before it starts clicking things with confidence.

exploring

reliable agent actions on live interfaces.

structured browser context that keeps useful meaning without carrying every useless pixel with it.

network systems that resume after disconnection, because connectivity is apparently optional now.

learning

browser internals.

systems design.

machine learning infrastructure.

thinking about

context as an interface problem.

how agents should know when not to click something.

software that survives imperfect conditions and imperfect humans.