Generated banner for Omniverse Realtime Viewer
Generated logo for Omniverse Realtime Viewer

Omniverse Realtime Viewer

OpenAIOfficialCreated by NVIDIA OmniverseDistributed by OpenAI

  • Skill
  • Meta (authoring, orchestration, planning, memory)
  • Software development
  • Apache-2.0
  • Medium risk

From the author

Use as the top-level router for Omniverse Realtime Viewer USD app requests and focused viewer reference documents.
— the description by NVIDIA Omniverse

AutoGineer summary

Written by AutoGineer from the skill’s own files, not by the author

The Omniverse Realtime Viewer skill acts as a central orchestrator for developing USD-based viewer applications. It provides a structured approach to selecting implementation recipes, enforcing architectural conventions, and managing dependencies. By utilizing a set of focused reference documents, it guides developers through the lifecycle of building, configuring, and validating viewer applications, ensuring consistent behavior across different delivery paths such as local workstations or streaming environments.

What it does

  • Routes development requests to specific implementation recipes based on the target delivery path.
  • Enforces architectural rules for USD rendering, stage management, and UI integration.
  • Provides guidance on camera selection, input handling, and viewport interaction.
  • Standardizes the integration of AOV switching and render product configuration.
  • Facilitates the capture of validation evidence for viewer readiness.
  • Maintains separation between user-provided USD files and application-specific session state.

When to use it

  • When initiating the development of a new USD viewer application within the Omniverse ecosystem.
  • When determining the appropriate delivery path for a viewer, such as local native or streaming.
  • When needing to implement consistent camera, selection, or UI behavior across viewer projects.
  • When integrating advanced rendering features like AOV switching into an existing viewer.

Advantages

  • Ensures architectural consistency across diverse viewer implementations.
  • Provides a clear, modular path for adding specific capabilities like input routing or data display.
  • Promotes clean separation of concerns between stage data and viewer-owned state.
  • Standardizes the validation process for consistent delivery.

Usage examples

“I need to build a browser-based viewer for USD files.”
The agent will route you to the streaming viewer recipe, providing guidance on WebRTC stream setup, UI integration, and viewport interaction.
“How do I implement AOV switching in my local USD viewer?”
The agent will provide the AOV switching reference, detailing how to map render vars and update the display state on the server.

Install

  1. Download the skill as a ZIP:

    https://mp-staging.autogineer.com/agent-skills/openai/omniverse-realtime-viewer/download
  2. Unzip it into the cross-client skills folder of your project:

    mkdir -p .agents/skills && unzip omniverse-realtime-viewer.zip -d .agents/skills

Documentation by NVIDIA Omniverse

<!-- SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. -->

<!-- SPDX-License-Identifier: Apache-2.0 -->

Omniverse Realtime Viewer

This is the top-level entry point for the Omniverse Realtime Viewer skill package. It is self-contained: all required routing, conventions, and validation guidance live in the selected references.

Use the focused reference documents as implementation recipes. This file chooses the right recipes and preserves the architectural rules that must hold across all generated viewer apps.

Instructions

Start by classifying the requested viewer, then read only the references needed for that delivery path and feature set. Implement the render path first, layer interaction and UI behavior on top of it, and finish by capturing validation evidence from references/validation.md.

Read Order

  1. Read references/routing.md to choose the delivery path and focused references.

  2. Read references/conventions.md before implementing camera, input, selection, viewport, streaming protocol, scene loading, or environment behavior.

  3. For broad viewer requests, read references/usd-viewer-app/README.md.

  4. If the delivery path is unclear, read references/streaming-vs-local/README.md.

  5. If the prompt includes layout, panels, controls, inspectors, status, or UX, read references/viewer-ux-workflow/README.md and then the focused viewer UI references. This applies to React/WebRTC, Tauri, Electron, ovui, ovwidgets, and Dear ImGui apps; "frontend" means user-facing UI, not only browser UI.

  6. For viewport interaction, read references/viewer-input-routing/README.md before references/camera-controls/README.md, references/native-picking-selection/README.md, or references/object-selection/README.md.

  7. Read only the focused capability references needed for the requested app.

  8. Use references/validation.md to capture review evidence before handoff.

Non-Negotiables

  • Use ovrtx for all USD and 3D rendering.

  • Browser apps display an ovstream WebRTC video stream plus UI. The browser does not render USD geometry.

  • Do not substitute WebGL, Three.js, Babylon.js, PlayCanvas, A-Frame, model-viewer, react-three-fiber, glTF browser viewers, or other client-side 3D renderers.

  • If local validation cannot run because the GPU/runtime environment is absent, scaffold the ovrtx path and document the runtime requirement. Do not add a browser-renderer fallback.

  • Keep user USD files unmodified. Viewer cameras, render products, render vars, settings, selection metadata, and runtime state belong in session/composite layers or app state.

  • Keep one owner for renderer.step(), stage mutation, native picking, selection writes, and live attribute writes.

  • Keep dependency acquisition in references/dependencies/README.md and deployment choices in references/cloud-deployment/README.md; do not duplicate package locations or deployment setup.

Focused Reference Families

  • Entry points and recipes: references/usd-viewer-app/README.md, references/streaming-viewer-recipe/README.md, references/ovui-local-viewer-recipe/README.md, references/streaming-vs-local/README.md, references/electron-shm-viewer/README.md, references/ovwidgets-editor-shell/README.md.

  • Rendering and stage: references/ovrtx-rendering/README.md, references/stage-loading/README.md, references/stage-management/README.md, references/render-settings/README.md, references/aov-switching/README.md, references/stage-hierarchy/README.md, references/stage-queries/README.md, references/stage-attribute-reads/README.md, references/prim-transform-safety/README.md, references/usd-sample-data/README.md.

  • Delivery and runtime: references/streaming-server/README.md, references/streaming-client/README.md, references/streaming-messages/README.md, references/streaming-lifecycle/README.md, references/local-viewer/README.md, references/tauri-local-viewer/README.md, references/cpp-native-viewer/README.md, references/headless-shm-cli/README.md, references/viewer-backend-interface/README.md, references/webgl-shm-transport/README.md.

  • Viewer UI/UX: references/viewer-ux-workflow/README.md, references/viewer-layout-patterns/README.md, references/viewer-control-patterns/README.md, references/viewer-data-view-patterns/README.md, references/viewer-feedback-status/README.md.

  • Interaction: references/viewer-input-routing/README.md, references/camera-controls/README.md, references/object-selection/README.md, references/native-picking-selection/README.md, references/selection-feedback/README.md, references/selection-animation/README.md, references/transform-manipulator/README.md, references/gl-viewport-overlay/README.md, references/ovui-library/README.md, references/prim-pick-effects/README.md, references/prim-info-display/README.md, references/viewport-overlays/README.md.

  • Infrastructure: references/dependencies/README.md, references/windows-native-setup/README.md, references/cloud-assets/README.md, references/cloud-deployment/README.md, references/troubleshooting/README.md.

Build Workflow

  1. Classify the prompt by delivery path, target user, required capabilities, runtime environment, validation needs, and explicit constraints.

  2. Select a small reference set. Start with the recipe or routing reference, then add focused capabilities such as camera, picking, hierarchy, properties, render settings, transform tools, cloud assets, or deployment.

  3. Read selected references before writing app code. Follow their build order, import order, data-channel contracts, and renderer ownership rules.

  4. Implement the core render path first, then input routing and camera, then selection and data panels, then scene/settings features, then packaging or deployment.

  5. Treat the selected references as the behavior contract for API shape, compatibility, and generated project structure.

  6. Capture validation evidence before calling the viewer ready.

Examples

  • For a browser viewer request, use the streaming recipe references plus camera, picking, hierarchy, properties, render settings, and stream-status references.

  • For a local workstation viewer request, use the local or native delivery references plus renderer setup, stage loading, viewport input, and validation.

Completion Checklist

  • Selected references match the user's intent and delivery path.

  • No code path uses a browser-side 3D renderer for USD.

  • The generated app has one clear owner for render stepping and stage mutation.

  • User USD files remain untouched by viewer-owned session data.

  • Camera, input, selection, scene loading, and stream behavior follow references/conventions.md.

  • Setup/build/run results and visual interaction evidence are captured with references/validation.md.

Files

references/streaming-messages/README.md

Streaming Messages

Triggers

Use this skill for custom streaming message, data channel, event_type, openStageRequest, getChildrenRequest, getPropertiesRequest, getPrimCountRequest, getStatsRequest, selectPrimsRequest, setViewportInputActive, changeAOVRequest, activeAOVState, availableAOVsResult, variants, or message protocol.

Application messages use this envelope in both directions:

{"event_type": "<MessageType>", "payload": {}}

Input events are not JSON messages. WebRTC input arrives through NVST's native input channel as binary InputEvent structs and reaches streaming-server on_input. SHM Python clients must use ovstream.ShmClient.send_input_event(); C clients use ovstream_shm_client_send_input_event(). Do not use JSON mouseInput. In-process clients should call the local Python/C++ APIs directly. Read viewer-input-routing for button normalization, viewport ownership, and click-vs-drag dispatch.

ovstream Callback Split

Keep the StreamSDK callback responsibilities separate:

CallbackPayloadUse it for
Server on_messageJSON strings, bytes, dicts, or the browser library wrapper {messageType,data}App protocol messages: stage switching, tree requests, property requests, prim count, AOV changes, variants, settings, and UI state queries.
Browser onCustomEventParsed app events from the serverReact state updates: openStageResult, getChildrenResult, stageSelectionChanged, availableAOVsResult, activeAOVState, and errors.
Server on_inputRaw ovstream.InputEvent objectsMouse, keyboard, wheel, and gamepad input for camera orbit/pan/zoom and click-to-pick.
Server on_unicodeComposed text inputIME, on-screen keyboard, paste, and other text events when a viewer needs text entry.

The browser does not implement camera math and must not forward pointer movement as app JSON. The WebRTC streaming library forwards raw input through NVST/ovstream; the Python server handles orbit, pan, zoom, drag-threshold click detection, and picking.

Golden-style server routing:

stream_server.on_connection = message_handler.on_connection
stream_server.on_message = message_handler.on_message
stream_server.on_input = message_handler.on_input
# Optional for composed text input:
# stream_server.on_unicode = message_handler.on_unicode
def on_input(self, event):
    import ovstream
    if event.type == ovstream.InputEventType.MOUSE:
        mouse = event.mouse
        if mouse.type == ovstream.MouseEventType.MOVE:
            server.camera.on_mouse_move(mouse.x, mouse.y)
        elif mouse.type == ovstream.MouseEventType.BUTTON:
            button = camera_button_from_ovstream(mouse.data, ovstream)
            if button is None:
                return
            is_down = mouse.button_state == ovstream.KeyState.DOWN
            if is_down:
                server.camera.on_mouse_button_down(mouse.x, mouse.y, button)
            else:
                was_click = server.camera.on_mouse_button_up(mouse.x, mouse.y, button)
                if button == 0 and was_click:
                    self._handle_click(mouse.x, mouse.y)
        elif mouse.type == ovstream.MouseEventType.WHEEL:
            server.camera.on_scroll(mouse.scroll_y or mouse.data)

Message Reference

FlowClient sendsServer sendsHandler responsibility
Open stageopenStageRequest {url}openStageResult {url,result,error?,root_prim_path?}Server loads USD into pxr worker and ovrtx, resets selection/highlight/AOV state, pushes root hierarchy
HierarchygetChildrenRequest {prim_path,filters?}getChildrenResult {prim_path,children}USD worker lists direct children and type/expandable metadata
PropertiesgetPropertiesRequest {prim_path,max_bytes?}getPropertiesResponse {prim_path,properties,truncated?}USD worker serializes attributes, relationships, metadata, variants, bounds summary
Prim countgetPrimCountRequest {}getPrimCountResult {count}USD worker traverses stage and returns total prim count
StatsgetStatsRequest {}getStatsResult {fps,latency_ms,...}Prefer client-side real WebRTC stats; server result may be placeholder
SelectionselectPrimsRequest {paths}stageSelectionChanged {prims}Server applies selected paths, updates highlight, and pushes canonical selection
SelectablemakePrimsSelectable {paths} or makePrimsPickable {paths}no required responseServer marks paths pickable/selectable
ResetresetStageRequest {} or resetStage {}optional openStageResult/loading messagesServer force-reloads the current stage; use force to bypass same-path skip
VariantsgetVariantsRequest {prim_path}getVariantsResponse {prim_path,variants}USD worker lists variant sets/options/current selection
Set variantsetVariantRequest {prim_path,variant_set,variant_selection}updated getVariantsResponse and/or hierarchy refreshServer applies variant, refreshes affected data
LoadingloadingStateQuery {}loadingStateResponse {url,loading_state}Server reports current load state
ProgressnoneupdateProgressAmount {amount}, updateProgressActivity {activity}Server pushes long load progress
AOV changechangeAOVRequest {aov}activeAOVState {active,available,result?,previous?,requested?,reason?}Server switches the render var copied into the video stream
AOV querygetAvailableAOVs {}availableAOVsResult {aovs,available}Server returns runtime-discovered displayable AOVs
Viewport inputsetViewportInputActive {active}no required responseServer gates native WebRTC input so DOM controls do not drive camera or picking
Render/settingstoggleSegView, setCameraGizmo, viewer-specific settingsimplementation-specific result or state pushServer updates renderer/view state

Exact event names matter. Current apps route getChildrenResult and getPropertiesResponse; older notes may say getChildrenResponse or getPropertiesResult. Accept old aliases when practical, but emit and document getPropertiesResponse for selected-prim properties.

getPropertiesResponse.prim_path is the response correlation key. Browser inspectors should compare it against the current selected prim path stored in a ref or resolver map, not against stale React state captured when the message handler was registered.

Payload Shapes

{"event_type":"openStageResult","payload":{"url":"samples/samples_data/stage01.usd","result":"success","root_prim_path":"/World"}}
{"event_type":"getChildrenResult","payload":{"prim_path":"/World","children":[{"name":"Cube","path":"/World/Cube","type":"geom","children":true},{"name":"Light","path":"/World/Light","type":"light","children":false}]}}
{"event_type":"getPropertiesResponse","payload":{"prim_path":"/World/Cube","properties":{"typeName":"Mesh","visibility":"inherited","xformOp:translate":[0,1,0]},"truncated":false}}
{"event_type":"getVariantsResponse","payload":{"prim_path":"/World/Car","variants":{"color":{"options":["red","blue"],"selection":"red"}}}}
{"event_type":"changeAOVRequest","payload":{"aov":"NormalSD"}}
{"event_type":"activeAOVState","payload":{"active":"NormalSD","available":["LdrColor","HdrColor","NormalSD","InstanceSegmentationSD","SemanticSegmentationSD","DepthSD","DiffuseAlbedoSD"],"result":"success","previous":"LdrColor"}}
{"event_type":"availableAOVsResult","payload":{"aovs":["LdrColor","HdrColor","NormalSD","InstanceSegmentationSD","SemanticSegmentationSD","DepthSD","DiffuseAlbedoSD"],"available":["LdrColor","HdrColor","NormalSD","InstanceSegmentationSD","SemanticSegmentationSD","DepthSD","DiffuseAlbedoSD"]}}

paths: [] in selectPrimsRequest clears selection.

{"event_type":"setViewportInputActive","payload":{"active":false}}

Use setViewportInputActive only as an app UI ownership hint. Mouse, keyboard, wheel, and touch events still travel through the native input channel, not JSON. The server should cancel active camera gestures when the flag changes to false.

Server Handler Map

Detailed server handler dispatch guidance lives in server-handler-map.md.

Selection Sync

Selection is bidirectional:

  • Viewport clicks update server selection, highlight, animation, server-side overlays, and then broadcast stageSelectionChanged.

  • Tree clicks must send selectPrimsRequest {paths} to the server; local React state alone does not update RTX highlight or animation.

  • The server's selectPrimsRequest handler should perform the same selection side effects as the pick path before broadcasting canonical selection.

def _handle_select_prims(self, payload: Dict[str, Any]) -> None:
    paths = [p for p in payload.get("paths", []) if isinstance(p, str)]
    prev_selected = set(self.server.selected_prims)
    new_selected = set(paths)
    self.server.selected_prims = paths

    if self.server._highlight_mgr:
        self.server._highlight_mgr.update_selection(new_selected, prev_selected)
    if self.server._animator:
        for path in new_selected - prev_selected:
            self.server._animator.select(path)
        for path in prev_selected - new_selected:
            self.server._animator.deselect(path)

    self.send_message("stageSelectionChanged", {"prims": self.server.selected_prims})

On the frontend, StageTree row selection should call:

sendMessage({ event_type: 'selectPrimsRequest', payload: { paths: selectedPaths } });

Dynamic Root Prim

Do not assume every stage uses /World. The reference server asks the pxr worker for a hierarchy root after loading:

def cmd_get_root_prim_path() -> Dict[str, Any]:
    if not _stage:
        return {"ok": False, "error": "no stage loaded"}
    world = _stage.GetPrimAtPath("/World")
    if world.IsValid():
        return {"ok": True, "path": "/World"}

    default_prim = _stage.GetDefaultPrim()
    if default_prim and default_prim.IsValid():
        return {"ok": True, "path": str(default_prim.GetPath())}

    for child in _stage.GetPseudoRoot().GetChildren():
        return {"ok": True, "path": str(child.GetPath())}

    return {"ok": True, "path": "/"}

Cache the result on the server as current_stage_root_path, include it in openStageResult, and use it for the initial getChildrenResult push:

self.send_message("openStageResult", {
    "url": active_url,
    "result": "success",
    "root_prim_path": root_path,
})
children = server._pxr.get_children(root_path)
self.send_message("getChildrenResult", {"prim_path": root_path, "children": children})

Stage Reload Semantics

Open-stage requests should be idempotent. Normalize paths before deciding whether a request points at the already loaded stage:

if not force and self.current_stage_url:
    requested_key = os.path.normcase(os.path.abspath(url))
    current_key = os.path.normcase(os.path.abspath(self.current_stage_url))
    if requested_key == current_key:
        logger.info("Stage already loaded, skipping reload: %s", url)
        return True

For explicit reset/reload messages, call the load path with force=True so a same-path reload is not optimized away:

def _handle_reset_stage(self, payload: Dict[str, Any]) -> None:
    if server.current_stage_url:
        server._load_stage(server.current_stage_url, force=True)

AOV Messages

AOV state is synchronized through normal data-channel messages; the video stream itself stays unchanged.

def _send_aov_state(self, extra: Optional[Dict[str, Any]] = None) -> None:
    available = server.get_available_aovs()
    active = getattr(server, "_active_aov", "LdrColor")
    payload = {"active": active, "available": available}
    if extra:
        payload.update(extra)
    self.send_message("activeAOVState", payload)
    self.send_message("availableAOVsResult", {"aovs": available, "available": available})
def _handle_change_aov(self, payload: Dict[str, Any]) -> None:
    requested = payload.get("aov") or payload.get("name")
    if not isinstance(requested, str) or not requested:
        self._send_aov_state({"result": "error", "reason": "Missing AOV name"})
        return

    previous = getattr(server, "_active_aov", "LdrColor")
    if server.set_active_aov(requested):
        self._send_aov_state({"result": "success", "previous": previous})
        return

    self._send_aov_state({
        "result": "error",
        "requested": requested,
        "reason": "AOV is not available for the current render product",
    })

toggleSegView should remain as a compatibility shim. Map it to InstanceSegmentationSD when enabled and LdrColor when disabled, then send normal AOV state.

On the frontend, accept both AOV event shapes:

case 'activeAOVState': {
  const payload = event.payload as ActiveAOVStatePayload;
  if (Array.isArray(payload.available) && payload.available.length > 0) {
    setAvailableAOVs(payload.available);
  }
  setActiveAOV(payload.active || 'LdrColor');
  break;
}
case 'availableAOVsResult': {
  const payload = event.payload as AvailableAOVsResultPayload;
  const names = payload.aovs || payload.available || [];
  if (names.length > 0) {
    setAvailableAOVs(names);
  }
  break;
}

Data-Channel Size Limit

Some WebRTC data-channel paths fail above roughly 65535 bytes per message. getPropertiesResponse is the common offender on complex prims.

Cap payloads before sending:

MAX_MESSAGE_BYTES = 60000

def capped_event(event_type: str, payload: dict) -> tuple[dict, bool]:
    encoded = json.dumps({"event_type": event_type, "payload": payload}, default=str)
    if len(encoded.encode("utf-8")) <= MAX_MESSAGE_BYTES:
        return payload, False
    if "properties" in payload:
        trimmed = {}
        omitted = 0
        for key, value in payload["properties"].items():
            candidate = {**payload, "properties": {**trimmed, key: value}}
            size = len(json.dumps({"event_type": event_type, "payload": candidate}, default=str).encode("utf-8"))
            if size > MAX_MESSAGE_BYTES:
                omitted += 1
            else:
                trimmed[key] = value
        return {**payload, "properties": trimmed, "truncated": True, "omitted_count": omitted}, True
    return {**payload, "truncated": True, "error": "payload too large"}, True

Prefer a capped single response over chunking unless the frontend already supports chunk assembly. If adding pagination, make it opt-in with request fields such as max_bytes, offset, or cursor, and keep the original response shape for older clients.

For selected-prim property panels, avoid sending full mesh buffers in the first place. Serialize array attributes such as points, normals, faceVertexIndices, faceVertexCounts, and primvars:st* as {length, preview, truncated} summaries unless the user explicitly requests a full geometry dump or paginated array viewer.

Adding A Message

If a project has server/config.py, add a message constant there; otherwise a literal string is acceptable. If generating code files, include these parts:

  1. server/message_handler.py: handler function, dictionary entry, payload validation, send helper call.

  2. USD worker module: a pure query/mutation function when the handler needs stage data.

  3. frontend/src/types/usd.ts: TypeScript payload interfaces and discriminated event type.

  4. frontend/src/App.tsx or the relevant component: sendMessage({ event_type, payload }) and response routing in onCustomEvent.

def _handle_my_feature(self, payload):
    result = self._do_something(payload.get("some_param", ""))
    self._send_message("myFeatureResponse", {"result": result})
sendMessage({ event_type: 'myFeatureRequest', payload: { some_param: 'value' } });

Backward Compatibility Rules

  • Never change the outer {event_type,payload} envelope.

  • Add optional payload fields; do not rename required fields in place.

  • Accept known aliases (makePrimsPickable/makePrimsSelectable, resetStage/resetStageRequest, aov/name) and normalize internally.

  • Unknown request fields should be ignored, not treated as fatal.

  • Unknown event types should log a warning and return an error response only if the frontend expects one.

  • Keep request/response names stable across server and frontend; verify the active onCustomEvent router before changing names.

  • Send current state on connect for browser clients that attach after startup.

See also: aov-switching, streaming-client, streaming-server, streaming-lifecycle, viewer-input-routing, stage-hierarchy, stage-management.

Requirements and compatibility

Written for: OpenAI Codex

Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta

“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.

The author says: Orchestrator skill. Downstream focused references may require NVIDIA GPUs, ovrtx, ovstream, ovui, OpenUSD, Python, Node/React, Tauri, Electron, C++, or cloud GPU deployment access depending on the selected viewer path.

Security

What our scanner noticed· Medium risk

  • Large base64-encoded blobs
  • Instructions that use the network

A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.

Attribution and licence

Created by
NVIDIA Omniverse
Publisher
OpenAI (Official)
Distributed by
OpenAI
Source
openai/plugins · plugins/nvidia/skills/omniverse-realtime-viewer · commit d420d13
Licence
Apache License 2.0 (redistribution allowed)
  • LICENSE (written by AutoGineer)
  • ATTRIBUTION.md (written by AutoGineer)
Synced

AutoGineer is not affiliated with or endorsed by NVIDIA Omniverse or OpenAI.

Download ZIP

The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download omniverse-realtime-viewer.zip