rtrvr-ai/rover

Leverages direct DOM and accessibility tree parsing instead of vision models, enabling millisecond-latency task execution entirely in-browser with zero remote infrastructure. Exposes a `POST /v1/tasks` API for AI agent handoffs and optional WebMCP task/tools discovery, with RoverBook providing agent analytics, session replays, memory persistence, and experiment attribution across repeat visits. Supports deployment across websites, Chrome extensions, Electron apps, and other browser-native environments via both script-tag embedding and npm SDK with SSR/SPA lifecycle control.

Available on npm.