Give it a task in plain English. It plans with a 120B model, executes in a real browser, and streams every action live to your screen.
Six specialised NVIDIA models — planner, executor, vision, embeddings, reranking, and fallback — working in concert.
Full Playwright automation — click, type, scroll, navigate, screenshot — on any public website.
Every action streams to your browser in real time via WebSocket. Watch the agent work step by step.
Embedding-based memory recalls past task results so the agent improves over time.
After each task the agent writes a plain-English summary of what it found and did.
Screenshots are interpreted by the vision model to verify state and recover from errors.