Canvas

What the canvas is for

The canvas is a free-form spatial surface in the center panel. It's where the agent places things that don't belong in a linear chat: charts, diagrams, code snippets, notes, images, tables.

Think of it as a whiteboard the agent maintains for you. As your session progresses, the canvas fills up with the artifacts of your work.

Elements

The canvas supports these element types:

| Element | Description |

|---|---|

| Chart | Metric plots, loss curves, bar charts. Auto-generated from experiments. |

| Code | Syntax-highlighted code blocks. Editable in place. |

| Note | Markdown text. Good for summaries, observations, TODOs. |

| Image | PNGs, JPGs from the pod filesystem. Generated plots, architecture diagrams. |

| Table | Data tables. Sortable columns, filterable rows. |

| Diff | Side-by-side code or config diffs. |

Placing elements

The agent places elements automatically as it works. When it generates a chart, it goes on the canvas. When it writes a summary, it goes on the canvas.

You can also place elements manually. Right-click the canvas and choose what to add. Drag elements to reposition them. Resize by dragging corners.

Agent-driven updates

Elements are live. If the agent is running a training loop and logging metrics, the chart on the canvas updates in real time. If you ask the agent to modify a code block, the canvas element updates to show the new version.

Navigation

| Shortcut | Action |

|---|---|

| Scroll / trackpad | Pan |

| Cmd+scroll | Zoom |

| Cmd+0 | Reset zoom to 100% |

| Cmd+Shift+F | Fit all elements in view |

| Click + drag (background) | Pan |

| Click + drag (element) | Move element |

Exporting

Right-click the canvas and select Export to download the entire canvas as a PNG, PDF, or JSON. Individual elements can be exported too. Right-click any element for options.