Skip to main content

Requirements

Argus runs on hardware at your site. This page lists what that hardware needs before you install Argus.

The hardware a site needs depends on its cameras. We size it with you. To plan a site, contact us.

How a site is laid out​

An Argus site has one site node and one or more edge nodes.

  • The site node runs the site database, the event streams, clip storage, the AI models for captions and Copilot, the site services, the dashboard and the HTTPS entrypoint.
  • An edge node reads camera streams, detects and tracks people and objects, and cuts a clip when an alert fires.

The default install runs both on one host: the site node plus one edge node named edge-1. To spread cameras over more hardware, add edge nodes on other hosts. Each camera is assigned to one edge node.

These are the containers on the site node:

ContainerWhat it does
site-postgresSite database: PostgreSQL 16 with the pgvector extension.
site-redisRedis. Carries detection events and clip messages between the services.
site-seaweedfsSeaweedFS, S3-compatible storage for alert clips.
site-ollamaOllama. Runs the vision model for captions and the language model for Copilot.
site-migrateRuns once on every start. Brings the site database schema up to date, then exits.
site-edge-nodeThe built-in edge node, edge-1. Detection (YOLO), tracking, clip cutting and the live view stream.
site-consumerWrites detections to the site database.
site-rule-engineChecks detections against your rules and raises alerts.
site-clip-updaterLinks each finished clip to its alert.
site-vlm-captionerWrites a caption for each alert clip.
site-retentionDeletes events and clips that are past your retention window.
site-apiThe dashboard, the site API and Copilot.
site-cloud-syncOptional status link. Idle unless you give it a site key. See Standalone sites.
site-caddyCaddy, the HTTPS entrypoint. The only container with published ports: 443, and 80 for the redirect to HTTPS.
note

The Compose file gives each container a fixed name. Run one Argus site per Docker host.

Operating system and software​

The site node and every edge node need:

ItemRequirement
Operating systemLinux.
ContainersDocker Engine with the Docker Compose v2 plugin, so that docker compose works.
Internet accessDuring install, to build the images and download the models. For a site without internet access, see Offline sites. Argus does not need internet access to run.
Ports443 and 80 free on the site node. You can move them with HTTPS_PORT and HTTP_PORT.

To use the dashboard, you need a current web browser that can play H.264 MP4 video. Alert clips are H.264 MP4 files. The dashboard is served over HTTPS. With the default certificate, each browser needs to trust the site's own certificate authority once. See HTTPS.

GPU (optional)​

Argus runs on CPU by default, which is enough to try it out. For a site in production, plan an NVIDIA GPU host, and contact us to size it.

On a host with an NVIDIA GPU, you add the docker-compose.gpu.yml file at install. It gives Ollama and the edge node access to all GPUs on the host.

For a GPU host you need:

  • An NVIDIA GPU with the NVIDIA driver installed.
  • The NVIDIA Container Toolkit, so that Docker can pass the GPU to containers.

To check that Docker can use the GPU, run:

docker run --rm --gpus all nvidia/cuda:12.4.1-base-ubuntu22.04 nvidia-smi

If the output lists your GPU, Docker can use it.

Without a GPU, captions and Copilot answers take longer, and an edge node can handle fewer cameras. See GPU and CPU for details.

Sizing​

Argus doesn't publish fixed hardware sizes. What a site needs depends on the number of cameras, their resolution and frame rate, how much happens in front of them and how long you keep clips. Contact us to size a site.

These defaults affect how much a site uses:

SettingDefaultWhat it affects
Detection modelYOLOv8nEdge node CPU or GPU use.
Frames with detections sent per cameraUp to 5 per second (FPS_LIMIT)Database growth and event traffic to the site node.
Clip buffer per cameraLast 15 seconds of video (BUFFER_SECONDS)Edge node memory.
Clip length5 seconds before and 5 seconds after the triggerClip storage.
Caption and Copilot modelsgemma3:4b, a local model in Ollama. Set with VISION_MODEL and COPILOT_MODEL.Site node CPU or GPU and memory use.
Search embedding modelnomic-embed-textSite node CPU or GPU use.
Retention30 days, checked every 6 hoursDatabase and clip storage.

Keep these points in mind:

  • The edge node runs detection on as many frames as it can keep up with, up to the camera's frame rate. FPS_LIMIT only limits how many frames per second it sends detections from. Each of those frames sends one event for every tracked person or object. A busy scene sends many more events than a quiet one.
  • The edge node keeps the last 15 seconds of decoded video for each camera in memory, so that it can cut clips. Memory use grows with resolution and frame rate. A lower-resolution stream from the camera reduces CPU, GPU and memory use. Many cameras offer a second, smaller stream for this.
  • When the edge node and Ollama run on the same host, they share its CPU or GPU. As you add cameras, move detection to edge nodes on other hosts.
  • Plan disk for the site database, the clips in your retention window, the AI models and the container images. Argus stores each detection event, not only alerts.

Network​

Argus runs on your local network. The edge node and the site node both connect to your cameras. Edge nodes on other hosts connect to the site node. Browsers connect to the site node over HTTPS, on port 443, for the dashboard, clip playback and live video.

See Cameras and network for the ports and firewall rules.

Checklist​

Before you install, check that you have:

  • A Linux host for the site node with Docker Engine and Docker Compose v2.
  • For GPU hosts: the NVIDIA driver and the NVIDIA Container Toolkit, checked with nvidia-smi as above.
  • The Argus release from VoisX.
  • A decision on the HTTPS certificate: the site's own certificate authority (default), a public DNS name with an automatic certificate, or your own certificate. See Install.
  • Internet access on the host during install, or the files for an offline install.
  • The RTSP address of each camera, and a camera account that can view the stream.
  • Network access from the site node to the cameras.
Was this page helpful?
On this page
Talk to our teamQuestions about licensing or sizing VoisX Argus for your site? We can help.Book a site assessment →