The KAGAMI mark КАГАМИ
kagami.bg/academy · lesson · machine-readable viewUPDATED 2026-10-03
IDENTITY
module
GX10-04-73 · Video agent with NVIDIA VSS (Video Search and Summarization Blueprint)
series
GX10 (local AI server class: NVIDIA GB10, e.g. ASUS Ascent GX10 / DGX Spark)
level
Advanced
duration
about 1.5 to 2 h plus image downloads
prerequisites
GB10-class machine with DGX OS, Docker and Docker Compose in the supported version band, NVIDIA Container Toolkit, NGC CLI, an NGC personal API key, a Hugging Face read token, Git with Git LFS
trust_label
UPDATED 2026-10-03 (rewritten against the NVIDIA VSS documentation, pages last updated 2026-07-16, deployment package tag v3.2.1) · NOT TESTED (no GB10 machine available during the update; no command in this lesson was run by the authors)
versions
VSS Blueprint deployment package v3.2.1 · sample data resource dev-profile-sample-data:3.2.0 · DGX OS 7.4.0 and driver 580.95.05 for DGX Spark · NVIDIA Container Toolkit 1.17.8+ · Docker 28.3.3 or newer and older than 29.5.0 · Docker Compose v2.39.1+ · NGC CLI 4.10.0+ · default LLM NIM nvidia/nvidia-nemotron-nano-9b-v2 · default VLM NIM nvidia/cosmos3-nano-reasoner
language
human view: english edition · bulgarian edition: /academy/gx10/ (same file name)
previous / next
GX10 series index (neighbouring lessons not yet published)
PURPOSE

Deploy the base vision agent of the NVIDIA VSS Blueprint with Docker Compose through the profile script, upload a video through the agent UI, ask questions about it in plain language, and generate a Markdown and PDF report. The lesson explains what is deployed (agent, UI, video storage service, LLM and VLM inference microservices, tracing), what is documented for the GB10 class (base and alerts profiles; summarization and search profiles are not listed for it), and what has to be decided before real footage of people is processed (GDPR, retention, access). Everything is from the documentation; nothing was run by the authors.

KEY CONCEPTS
COMMANDS / PATHS
CHECKLIST
NEXT MODULE

GX10-04-120 Video archive search with AI (own pipeline with OpenCLIP and pgvector) · GX10-04-30 Semantic video search · GX10 series index: kagami.bg/en/academy/gx10/ · offer: Quick experiment (kagami.bg/stalbata/)

SOURCES
TAGS
gx10nvidia-gb10arm64nvidia-vssvision-agentvlmdocker-composevideo-analyticsgdpr
UPDATED · 03.10.2026

Video Agent with NVIDIA VSS on GX10

You upload a video recording, ask in plain language "is the worker wearing a hard hat?" and get an answer with a supporting frame — plus a ready report in Markdown and PDF. This is NVIDIA's ready-made video agent, brought up with one script, rather than one more home-built pipeline.

⏱ 1.5–2 h Advanced GX10 NVIDIA GB10 · 128 GB unified memory VSS Blueprint · Docker Compose
NVIDIA VSS Blueprint🔒 local Docker Compose🔒 local Remote LLM (optional)🌐 global
🔄
UPDATED · 03.10.2026 — what changed
The lesson was written anew against the NVIDIA VSS documentation (the pages were last updated on 16.07.2026, the deployment package is tagged v3.2.1). The old version described a home-built pipeline with other models and another vector database, and a table of timings "on GB10" without a source. We removed: that pipeline, the performance table (numbers with no measurement), the scenarios with specific customers and sites, and the addresses and code with internal paths. We added: what the base agent consists of, what the documentation says about the GB10 class, the exact Docker and driver requirements, the human in the loop for reports, a security note (VSS is meant for an isolated network) and a legal framework for recordings of people.
⚠️
What we have not run ourselves
During the check we had no machine of the GB10 class. Not one command in this lesson was run by us — they are checked against the documentation, which is why there is no "TESTED" label. Also unchecked: how long downloading and starting the images takes, how much memory the models take together on your machine, the quality of the answers on your own recordings and how it behaves with Bulgarian questions. NVIDIA's documentation puts DGX Spark machines first; the ASUS Ascent GX10 is of the same class, but NVIDIA's list does not name it explicitly — ask your supplier.

01What you'll learn

02Before you start

💡
What the documentation says about GB10
In the requirements table for DGX Spark the base and alerts profiles are listed with an external LLM. The summarization and search profiles are not listed for it. The documentation adds that a local LLM and VLM on one machine are supported, but an external LLM is recommended for a faster response, and fully local execution of all workflows is planned for a later release. That is why we show only the base agent below.
🛡️
Isolated network only
NVIDIA states it explicitly: deploy VSS in a trusted, isolated network and do not expose its services to untrusted networks. The release has no login, TLS or request limiting — they are expected from your infrastructure. So do not open the ports to the internet.

03Steps

  1. What you get

    VSS is a set of reference architectures. It has three layers: real-time video intelligence (object detection, embeddings, scene descriptions from a vision-language model), downstream analytics (behaviour, alert verification) and an agent that assembles all of that into reports, answers and search. Here we deploy the smallest piece — the base agent:

    PartJob
    VSS AgentTakes the request, decides which tools to call and assembles the answer. The recommended LLM is Nemotron Nano 9B v2
    Agent UIWeb chat with video upload and several views
    Video IO & Storage (VIOS)Ingests, records and plays back the videos for the agent
    LLM (NIM)Reasons and picks tools
    VLM — Cosmos3 Nano Reasoner (NIM)"Watches" the video and answers questions about it
    PhoenixTraces the agent's steps

    All parts can also be used on their own through REST and MCP; for this lesson we need only the chat.

  2. Prepare the machine

    First check the versions: docker version, docker compose version, ngc --version. If Docker is newer than 29.5.0 or older than 28.3.3, the images may fail to download. Then apply the settings from the requirements page: a file in /etc/sysctl.d with network parameters and a cache-cleaner script (for DGX Spark) that periodically drops the system cache, and have Docker use the cgroupfs cgroup driver. Take the exact text of the scripts from the requirements page — it changes, and nothing here has been run.

    💡
    Why the cache cleaner
    The memory of these machines is shared between the processor and the graphics chip; with large models the system cache can eat it up. That is why NVIDIA requires this script on DGX Spark. It works only while the session is alive — for permanence you make it a system service.
  3. Download the package and the sample recording

    The deployment package is in an NVIDIA repository on GitHub and uses Git LFS for large files. The documentation points to the tag v3.2.1.

    bash · not run
    git clone https://github.com/NVIDIA-AI-Blueprints/video-search-and-summarization.git
    cd video-search-and-summarization
    git checkout tags/v3.2.1
    git lfs install
    git lfs pull

    The sample recording comes from NGC. It is needed only for a trial; you can also use your own short recording, provided it shows no people who have not given consent.

    bash · not run
    ngc registry resource download-version nvidia/vss-developer/dev-profile-sample-data:3.2.0

    Before that, set up the NGC CLI with ngc config set and choose an organization — without one the resources cannot be downloaded.

  4. Set the keys and deploy

    The keys are set as environment variables in the same terminal you run the script from. Do not write them to a file and do not commit them to a repository.

    bash · not run
    export NGC_CLI_API_KEY='<ngc-api-key>'
    export HF_TOKEN='<hf-token>'
    
    # all options of the script
    deploy/docker/scripts/dev-profile.sh --help

    For DGX Spark the documentation shows two ways. The first — with an external LLM (its address is yours: another machine, or a local server on another port):

    bash · not run
    export LLM_ENDPOINT_URL=https://<llm-address>
    deploy/docker/scripts/dev-profile.sh up -p base -H DGX-SPARK --use-remote-llm

    The second — both the LLM and the VLM as containers on the same machine (according to the "edge" deployment guide):

    bash · not run
    deploy/docker/scripts/dev-profile.sh up -p base \
      --hardware-profile DGX-SPARK \
      --llm nvidia/NVIDIA-Nemotron-Nano-9B-v2-FP8 \
      --vlm nvidia/cosmos3-nano-reasoner
    ⚠️
    The two examples are not written the same way
    The first uses -H, the second --hardware-profile. We have not checked whether they are the same; run --help and use the form your version shows. The "edge" guide also has a variant with a smaller LLM (Nemotron 3 Nano 4B) with its own configuration file — we do not cover it here.

    The script downloads the containers and starts them. When it finishes, check:

    bash · not run
    docker ps
    curl -fsS http://<server-address>:8000/health

    The agent should be reachable on port 7777 (chat); 8000 is its REST interface and 30888 the video interface. If the page does not open, the documentation points to checking the machine's firewall: containers on a bridge network may not reach services on the host.

  5. Upload a recording and ask

    Open the chat on port 7777 from a browser in the same network. Drag the video into the chat window — the agent answers with playback. Then you ask, for example (the examples are from the documentation, for the sample recording):

    chat
    Is the worker wearing PPE?
    When did the worker climb up the ladder?
    List all available sensors
    Take a snapshot of <sensor-id> at <timestamp>

    The agent shows its intermediate steps before giving the answer. The documentation gives the examples in English; how it behaves in Bulgarian, we have not checked. The answer is the model's opinion, not evidence: for every important claim, look at the frame and the recording.

  6. A report — and the human in the loop

    Ask "Can you generate a report for this video?". Before it starts, the agent shows you the prompt for the VLM and waits. You have five options:

    • submit empty — you approve and a report is generated;
    • type a new prompt or edit the one shown;
    • /generate <description> — the LLM drafts a prompt from your description;
    • /refine <instructions> — the LLM adjusts the current one;
    • /cancel — you decline.

    This is a good practice, because the report is only as accurate as the prompt. Reports come in Markdown and PDF and sit on port 8000 under static. By default they are kept in memory — after a restart of the agent container they disappear. For permanent storage you have to turn on saving a local copy and mount a directory from the machine (described in the documentation for reports).

  7. People in the frame: the legal framework

    ⚖️
    Recording and analysing people has legal requirements
    Video surveillance, including automatic description of recordings with people and the reports made from them, is processing of personal data and falls under data-protection rules (GDPR) and national law. The legal basis, informing people, retention period, access to the recordings and reports, and the rights of the people concerned are questions that depend on the site and the purpose. Recommendation: before you run a system on real recordings, check with a lawyer and write down their decisions. This lesson is technical and is not legal advice.
    • Good technical practice: start with sample recordings without people; limit who has access to the chat and the reports; do not turn on recognition of specific individuals.
    • Keep recordings and reports as long as needed and no longer; set a period and apply it.
    • Do not expose the ports to the internet — the agent has no login of its own.
  8. Shut it down

    bash · not run
    deploy/docker/scripts/dev-profile.sh down

    The command stops and removes the agent's containers. When switching profile (for example to summarization), the documentation advises clearing the old data volumes as well.

04Check

Quiz

1. What kind of network does NVIDIA expect for a VSS deployment?

2. Which profiles are listed for DGX Spark in the requirements table?

3. Why does the agent show you the VLM prompt before it starts a report?

4. What is the most sensible thing to do before running the agent on recordings of real people?

05What's next

06Sources

  1. NVIDIA VSS · Introduction 🔒 local — architecture, three layers, isolated network.
  2. NVIDIA VSS · Prerequisites — hardware, versions, kernel settings, keys, GPU profiles.
  3. NVIDIA VSS · Quickstart — deploying the base agent, questions, report, shutdown.
  4. NVIDIA VSS · Edge Deployment — DGX Spark and local models.
  5. GitHub: NVIDIA-AI-Blueprints/video-search-and-summarization — the deployment package.
  6. build.nvidia.com · VSS Blueprint 🌐 global — project page.