Pilot

Pilot turns your browser into something an AI can actually operate — and turns the things you do by hand into reusable "skills" other agents can run. • Chat in the side panel: a local AI agent drives the tab you're on, showing its reasoning and tool calls in one collapsible "thinking" block. • Bring your own agent: connect a local AI coding agent you already use, or any custom command-line agent, with your own model or API key. • Record → skill: record clicks/typing/navigation (each with a cropped screenshot) and narrate with your voice; Pilot authors a skill and exports it to every agent installed on your machine. • Deterministic replay: a saved skill runs in one call, with versioning and a get/update tool so any agent can refine it. • On-page visuals: a cursor glides to the element, a glow marks it, and the page shows a "Pilot is controlling this tab" frame. • Auto-launch: if no browser is open, Pilot starts one when an agent needs the page. Permissions: • "debugger" — required to read the page's accessibility tree and send native input (the same browser debugging protocol used by Chrome DevTools). Can be switched to DOM mode in Settings. • "Read and change all your data on all websites" — required to act on whatever page you ask the agent to operate.