> ## Documentation Index > Fetch the complete documentation index at: https://docs.runchat.com/llms.txt > Use this file to discover all available pages before exploring further. # Scene > A shared 3D scene beside every workflow, where you, your apps and a team of agents can model at the same time Every saved workflow has a 3D scene alongside its canvas. The agent can model in it, you can bring in your own models, and [apps on the canvas](/concepts/editor/apps) can place what they make there. Changes appear live for everyone who has the workflow open, so several agents can build different parts of one site at once. The Scene view showing a terrain with buildings, trees and a cliff assembled by several apps ## Opening the scene Use the **Workflow / Scene** switch at the top of the editor. The canvas stays where it was underneath, so you can flip between the two at any time. In the Scene view: * **Orbit, pan and zoom** with the mouse, as in most 3D viewers * **Click** an object to select it, and **Shift-click** to add to the selection * **Double-click** an object to select it and zoom to it * **Fit** (bottom left) frames everything, and **Selection** frames what you have selected A new workflow's scene is empty until something is added to it. ## Modelling with the agent Open the Scene view and ask the agent for what you want: "block out a two-storey house on a 20 by 30 m site", "add a row of street trees along the north edge". The agent builds with simple shapes (boxes, cylinders, spheres and extruded outlines), small custom meshes, and hosted 3D models, then renders the scene to check its work before replying. Select objects before you send a message and they go to the agent with it, the same way selected nodes do on the canvas. That lets you point at things: "make this one three storeys", "move these behind the cliff". A building selected in the Scene view, outlined in blue, with a prompt in the chat about adding a pergola to it ### Units and orientation The scene is in **metres**, with **Z up**, X east and Y north. An object's position is the centre of its base, so a box at height 0 sits on the ground rather than half under it. Ask in real dimensions and the agent models to them. ### Building with a team With [agent teams](/concepts/agents/chat#agent-teams) on, the lead can hand parts of a scene to teammates. The **Modeller** specialist works in the scene by default. Each agent keeps its work in its own group and only changes objects it made or was handed, so a landscape, a building and the planting around it can go up at once without agents overwriting each other. Every object records who last changed it, so you can ask the agent "who placed the trees?" or "remove everything the modeller added". ## Adding your own models Click the **+** button in the Scene view: The Scene view's add menu with Import model and Capture view to canvas * **Import model (.glb, .obj)** uploads a model from your computer. OBJ files are converted to GLB. Each file can be up to 25 MB. * **Pick from Rhino** appears when the [Rhino plugin](/plugins/rhino) is connected. Select objects in Rhino and press Enter, and they come in at their real position and size, converted to metres. Surfaces, solids, SubD and meshes come across; curves, points and blocks are skipped. This needs Rhino 8. The agent can also import from a connected Rhino session, a selection or one layer at a time, so each part stays separately movable. Models added to a scene, by you, the agent or an app, show in your [Gallery](/concepts/dashboard/overview#gallery) under the **Scene** source filter, with a 3D thumbnail. ## From the scene to a render **Capture view to canvas** (in the same **+** menu) takes the scene as it is framed now and puts it on the canvas as an image input node. Connect that to an image model to turn a massing study into a render, in the same way as [rendering from a Rhino screenshot](/examples/rendering-from-rhino-screenshots). ## Apps that build in the scene [Apps](/concepts/editor/apps) on the canvas can add geometry to the scene too: a terrain generator, a building designer, a planting tool. Each one designs on its own, then places its result in the scene when you (or an agent) choose **Add to scene**. Apps can also read what is already there, so a building can sit on the terrain another app made. Three apps on the canvas, a terrain designer, a building designer and a plant designer, whose results make up the scene ## Saving and sharing The scene is saved and backed up with the workflow: restoring an earlier version brings back its scene, and copying the workflow copies the scene too. Anyone who can view the workflow can view the scene; only people who can edit it can change it. ## From the CLI and MCP The scene tools are available to external agents through the [MCP server](/concepts/integrations/mcp) and the [CLI](/concepts/integrations/cli): | Tool | What it does | | - | - | | `read_scene` | Lists every object with its id, kind, dimensions, who last changed it, and the box it occupies | | `edit_scene` | Adds, changes or removes objects, up to 200 edits per call | | `view_scene` | Renders the scene from an iso, top, front, back, left or right view, optionally framed on one object | ```bash theme={null} runchat read_scene --runchat_id runchat view_scene --runchat_id --view top ``` ## Limits * Up to 2,000 objects per scene * Hosted model files up to 25 MB each * Small custom meshes stored in the scene itself up to 2,000 vertices; anything larger is saved as a model file This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.