How GhostLine Works

Five panels that work together: a control bar for prompts, a viewer for 3D trajectories, replay controls, a research sidebar, and floating tools for settings and token inspection.

You type a prompt. The backend generates tokens one at a time. Each token arrives with its geometry, gets buffered so you can watch at human speed, then appears on screen.

Every generation is automatically recorded. You can save it, load it back later, scrub through it token by token, and inspect any moment — no backend needed.

The research sidebar sits right next to the viewer. It reads the same signals and sessions you're already watching — compare runs, test hypotheses, sweep parameters.

Switch between layout presets, save your workspace, and point at a local or remote GPU backend — all from the same interface.