# Build a room or table pack with an agent > An AI agent builds a DiceyTable room or table pack with room_set, table_pack_set and fields_set, then checks it with editor screenshots and a physics test. - Question: How does an agent build a room or a table pack? - Source: https://diceytable.com/docs/ai-agents/workflows/build-a-room-or-table - Last verified: 2026-10-09 An agent builds a room or table pack in the same editor as a game. It sets the room with `room_set` and `environment_set`, or the table with `table_pack_set` and a few `fields_set` calls. Then it checks the result with screenshots in the editor and a quick physics test. A **room pack** is the space around the table: walls, floor, windows, posters, scenery and lights. A **table pack** is the table itself: its model, its surface, its size, the shape pieces land on, and the ring of seats. Got a table model already? The page on [importing 3D models](https://diceytable.com/features/import-3d-models) shows what DiceyTable can take in. ## What you'll need - An editor tab [paired with your agent](https://diceytable.com/docs/ai-agents/setup/pairing). - A key with **Edit projects**, plus **Play at the table** for the physics test. - Any models or textures, ready to [import](https://diceytable.com/docs/ai-agents/workflows/importing-assets). Everything in [Build a game with an AI agent](https://diceytable.com/docs/ai-agents/workflows/build-a-game) about looking first, the project lease and saving applies here too. ## Step 1: Create or open the pack The agent calls `project_create` with `subject: "room"` or `subject: "table"` and a name. For an existing pack it uses `project_open`. The editor opens in the matching mode, and `page_info` reports it as the `subject`. ## Step 2: Room packs: decor, lights and sky - `room_set` sets the decor and lighting: surfaces, openings, posters, the view outside and scenery lights. - `environment_set` sets the sky, the exposure and the light that comes from the surroundings. - `fields_set` reaches everything else by field id. The `room.*` panels, such as `room.surfaces` and `room.light`, are in the [field catalogue](https://diceytable.com/docs/ai-agents/reference/fields). Room lights are ordinary entities with a `light` component, so `component_set` works on them too. Models and textures come in through `asset_import`, and `material_create` and `material_set` handle surfaces. [Room and lighting](https://diceytable.com/docs/editor/working/room-and-lighting) explains what each value does. ## Step 3: Table packs: model, surface, collider and seats - `table_pack_set` sets the table model, its surface material and its size. - `fields_set` on `table-pack.play-surface` sets where pieces rest: the play surface's height and size. - `fields_set` on `table-pack.collider` sets the **collider**, the invisible shape pieces land on. - `fields_set` on `table-pack.seat-ring` sets where the seats sit around the table. A collider that doesn't match the visible top is the usual reason dice float above the felt or sink into it. For a custom model, `asset_import` brings in the file and converts formats such as FBX and OBJ to GLB. The Model editor's collision and interior fields are under `item.model.collision` and `item.model.interior`. ## Step 4: Assign the pack to a game `pack_assign` sets which room or table a game uses. The agent passes `role: "room"` or `role: "table"` and the pack's id. A `packId` of `null` clears it, which means the platform default (Modern Room for a room, Square Table for a table). See [Assign a room or table](https://diceytable.com/docs/editor/workflows/assign-a-room-or-table). ## Why screenshot a room pack in the editor? > **Known gap.** Room decor does not render in a preview tab today. The capture code says so in its own warning > (`apps/web/src/playcanvas/photo/agentCapture.ts`, `ROOM_PACK_PREVIEW_CAPTURE_WARNING`: *"room-pack decor does not > render in the preview tab"*). A preview screenshot of a room pack shows the table in an empty space and comes back > with that warning. Table packs render normally in a preview, and the room looks right in the editor. So take room > pack screenshots in the editor viewport until this is fixed. > See [Known limitations](https://diceytable.com/docs/reference/known-limitations). In the editor, the agent asks for something like this: ```json { "camera": "iso", "hide": "chrome", "settle": true } ``` The editor viewport also lets the agent camera look from anywhere, even from under the table to check the bottom of a model. [Take screenshots with an agent](https://diceytable.com/docs/ai-agents/workflows/screenshots) lists the cameras. ## Step 5: Test that pieces behave A table pack is only right if things land on it properly. Open a game that uses it, and start a preview with `playtest_start`. Then the agent: 1. Drops a few dice and cards with `table_spawn`, rolls the dice with `table_roll`, and waits with `table_wait_settled`. 2. Checks with `table_assert` that everything is still on the table and at rest. 3. Reads the resting heights with `table_snapshot`. For a repeatable look at a fall, `table_step_physics` freezes the physics and steps it forward a set number of times. It only works in a private preview, never in a live room. [Play-test your game with an agent](https://diceytable.com/docs/ai-agents/workflows/playtest) covers these tools. ## Step 6: Save and publish The agent runs `project_save` and `project_validate`, then `publish_preview` and `publish_request`, the same as for a game. Credits for imported models and textures follow the same [review rule](https://diceytable.com/docs/ai-agents/concepts/ai-attested-credits): you check each one before you publish. ## Next steps - [Import models, images and sounds](https://diceytable.com/docs/ai-agents/workflows/importing-assets): bring in the table model and textures. - [Lighting a scene](https://diceytable.com/docs/editor/workflows/light-a-scene): make the room look its best. - [Get your game ready to publish](https://diceytable.com/docs/ai-agents/workflows/publish): the last step for any pack.