Setup VSS and Skills#

Minimum path to a deployed VSS stack driven from the NemoClaw / OpenClaw chat UI. This notebook and setup are intended to run on the same machine that hosts VSS, that is, a machine capable of running VSS. All steps run on that VSS-capable host (such as the Brev instance) through deploy/docker/scripts/deploy_nemoclaw_vss.ipynb. For the full variable reference and optional verification cells, see Additional Configuration.

Note

To run NemoClaw on a separate machine from the VSS host, refer to the NemoClaw for OpenClaw model card and the NemoClaw OpenClaw user guide.

Prerequisites#

Access to the VSS repository.
A Brev instance with Docker, curl, python3, and uv.
- Before launching, you can edit the Brev hardware profile and cloud service provider (CSP). See Brev hardware and CSP selection.
A Jupyter kernel with Python 3.11 or newer to run deploy_nemoclaw_vss.ipynb.
- Auto-fulfilled when running from the Brev launchable above.
NGC_CLI_API_KEY (NVIDIA legacy API key)
- for nvcr.io login, NGC CLI configuration, and NGC model artifact downloads.
An OpenClaw chat model provider:
- NVIDIA_API_KEY for the build.nvidia.com provider, OR
- COMPATIBLE_API_KEY for a cloud or local OpenAI-compatible endpoint.

Steps#

Clone the VSS repository. The VSS repository should be cloned at ~/video-search-and-summarization.
- If your checkout is elsewhere or the host has multiple clones, set VSS_REPO_DIR before launching Jupyter so the notebook resolves the intended checkout.
Open the notebook. deploy/docker/scripts/deploy_nemoclaw_vss.ipynb.
- If you are re-onboarding a host that previously ran an older OpenShell gateway, review the NemoClaw sandbox lifecycle guidance before running the notebook.
Set required values (Section 1.1):
```
NGC_CLI_API_KEY = "..."
HARDWARE_PROFILE = "RTXPRO6000BW"  # H100 | L40S | RTXPRO6000BW | DGX-SPARK | IGX-THOR | AGX-THOR | OTHER
```
Select the supported HARDWARE_PROFILE that matches your system specs. DGX-SPARK, IGX-THOR, and AGX-THOR are edge hardware profiles and are valid only for the base and alerts profiles.
Choose one OpenClaw model provider (Section 1.2). NemoClaw uses this model provider for its agent workflow.
- SOTA cloud model, recommended for best agent quality. Use any OpenAI-compatible cloud endpoint:
```
NEMOCLAW_ENDPOINT_URL = "https://<provider-base-url>/v1/"
NEMOCLAW_MODEL = "<model-id>"
COMPATIBLE_API_KEY = "<provider-api-key>"
```
- Local OpenAI-compatible model. Use a local or LAN model server that the sandbox can reach:
```
NEMOCLAW_ENDPOINT_URL = "http://host.openshell.internal:8000/v1"
NEMOCLAW_MODEL = "nvidia/nemotron-3-super-120b-a12b"
COMPATIBLE_API_KEY = "EMPTY"
```
  COMPATIBLE_API_KEY must be a non-empty value even when the local server ignores it. From inside the sandbox, the host is reached through host.openshell.internal.
- build.nvidia.com NVIDIA-hosted model. Clear the custom endpoint values and set NVIDIA_API_KEY:
```
NVIDIA_API_KEY = "nvapi-..."
NEMOCLAW_MODEL = "qwen/qwen3.5-122b-a10b"
```
  Leave NEMOCLAW_MODEL blank to use the installer default, nvidia/nemotron-3-super-120b-a12b.
Review advanced defaults (Section 1.3). These are the advanced knobs, which you usually do not need to change when running with the default settings.
Run Preflight (Section 2). Every required check should print OK. The preflight validates the NemoClaw init script, policy file, OpenClaw plugin, skills/ directory, MCP config/helper, host tools, and NGC_CLI_API_KEY.
Pin Docker (Section 2.1). Run this before installing NemoClaw. The cell pins the tested Docker stack: Docker CE 29.4.3, buildx 0.33.0, Compose 5.1.3, and containerd 2.2.3. A Docker downgrade restarts dockerd and would disrupt live sandbox containers if run later. It is safe to re-run.
Run Install (Section 3). The notebook installs or reuses the pinned NemoClaw release, then runs deploy/docker/scripts/nemoclaw/init_nemoclaw.sh. On completion:
- NemoClaw sandbox demo is created or reused.
- The OpenShell inference provider is configured for the selected build.nvidia.com, cloud OpenAI-compatible, or local OpenAI-compatible provider.
- The VSS policy preset is merged into the live sandbox policy.
- The NGC credential provider is configured and the NGC CLI is installed inside the sandbox on a best-effort basis.
- The .openclaw plugin is packed and installed into the sandbox, which installs the repository’s VSS skills and copies the NemoClaw workspace templates.
- The OpenClaw UI allowed origin, gateway, dashboard forward, and optional webhooks are refreshed.
Optionally verify the sandbox (Section 4). This checks sandbox state, active policy metadata, webhooks, plugin health, installed skills, and copied workspace files.
Prepare the host for local NIM-backed VSS profiles (Section 5):
- 5.1 - install and configure the NGC CLI on the host.
- 5.2 - authenticate Docker against nvcr.io.
- 5.3 - run uv sync under services/agent/.
These steps are required for local NIM-backed profiles and for profile flows that pull NGC model artifacts during docker_up.
Start the MCP server (Section 6.1). This launches the host-side VSS Orchestrator MCP server on port 9988 and polls vss_orchestrator__profiles until the server is healthy. The server log is written to <VSS_REPO_DIR>/.orchestrator-artifacts/vss_orchestrator_mcp.log.
Open the OpenClaw UI (Section 6.3). Section 6.2 is reference material for the chat-driven deployment flow and has no code cell of its own. Open the UI and smoke-test it before returning to the Section 6.2 prompts.

The Section 6.3 cell prints a fresh UI link with a gateway token. Open it in a browser.

The OpenClaw chat UI once the link is opened.#

Not on Brev? Open an SSH tunnel from your laptop before opening the UI:
```
ssh -L 18789:127.0.0.1:18789 <user>@<nemoclaw-host>
```
Smoke-test chat, skills, and tools from the OpenClaw UI:
```
hello
what model are you using?
list your available skills
show me the deployment tools
list the available VSS deployment profiles
```
The Skills tab should show the installed VSS skills.

Listing the installed VSS skills from the OpenClaw chat.#

The tools prompt should enumerate the nine vss_orchestrator__* tools.

Inspecting the available VSS orchestrator MCP tools from chat.#

Deploy VSS by chat (Section 6.2). Examples:

Prompt	Tools invoked
“List the available VSS deployment profiles.”	`profiles`
“Check that my host meets the prerequisites for the alerts profile.”	`prereqs`
“Deploy the VSS `alerts` profile in `verification` mode.”	`docker_generate` -> `docker_up` -> `docker_status`
“Fetch the last 20 lines of logs from `vss-alert-bridge`.”	`docker_logs`
“Tear down the VSS deployment.”	`docker_down` -> `docker_status`

Deploying the alerts profile from chat.#

Running the search profile from chat.#

For the complete tool reference, see VSS Orchestrator MCP server.

Add Video Sources#

Once the VSS stack is deployed, open the VSS UI and use the Video Management tab to upload video files (MP4, MKV) or add live RTSP camera streams for the agent to ingest. See Video Management Tab for the full walkthrough of upload, RTSP setup, and stream management.