Installation

System requirements

Choosing an image variant

Hardware acceleration for AI detection comes in 4 variants for different hardware. Picking the wrong variant only makes detection slower or non-functional; everything else (recording, playback, two-way audio) is unaffected. The one-line install script auto-detects your hardware and picks the right one for you — the table below is just a reference for those who want to choose manually.

Picking a tier by hardware (detailed reference)

When buying a new machine or evaluating an existing one, use the 5 tiers below grouped by CPU horsepower. Each lists the recommended image, the number of cameras it can reliably handle, and which AI detection features it can run smoothly (you still enable them manually in the web admin).

Detection features each tier can run smoothly

Higher tiers can reliably enable more features. On a low tier, turning on too many features makes inference fall behind the camera frame rate and miss events. The table below shows the detection capabilities recommended for each tier — ✅ means recommended, ❌ means not advised on that hardware. A fresh install enables only object detection by default; everything else must be turned on manually in web admin → Settings → Detection as needed.

One-line deploy (recommended)

Run a single command and the script auto-detects your hardware, picks the best-fit image variant, pulls the image, and starts the container — 5–15 minutes end to end depending on your network.

curl -fsSL https://cdn.yun-kan.com/yunkan-install.sh | bash

What the script does:

• Checks that the system is Linux x86_64 with Docker installed
• Checks that ports such as 23406 / 23880 / 24214 / 24215 / 23515 are free
• Probes hardware: NVIDIA GPU (including VRAM) / Intel iGPU / Intel NPU
• Automatically picks the best-fit image variant among cpu / openvino / cuda / trt
• Pulls the image (from Aliyun by default) + writes a docker-compose file + auto-generates a strong random secret
• Starts the container and waits for the health check to pass (about 60 seconds)
• Prints the browser URL (e.g. http://192.168.1.10:23406/)

Common flags

# Only show the recommended variant, don't deploy
curl -fsSL https://cdn.yun-kan.com/yunkan-install.sh | bash -s -- --detect-only

# Force the trt variant and skip prompts
curl -fsSL https://cdn.yun-kan.com/yunkan-install.sh | bash -s -- --variant trt -y

# Offline deploy (scp the image tar.gz to the machine first)
curl -fsSL https://cdn.yun-kan.com/yunkan-install.sh -o install.sh
bash install.sh --offline /path/to/skyview-image.tar.gz

Deploy by platform (pick your device)

The one-line script is easiest on a generic Linux + Docker environment, but NAS, HA OS, PVE and similar platforms each have their own container-management UI, network mode, iGPU passthrough method, and known pitfalls. Below are dedicated guides for common target platforms, including each platform's specific compose file, UI deployment path, firewall config, and FAQs:

• Synology DSM 7.2+ — Container Manager + Intel-iGPU models (DS920+ / DS423+ / DS224+, etc.)
• Unraid 6.12+ — Compose Manager + Intel GPU TOP / NVIDIA Driver plugin
• fnOS — Debian 12 base + built-in iGPU on N100/N305
• UGREEN UGOS Pro — x86 models (DXP2800/DXP4800/DXP6800, etc.)
• TrueNAS Scale 24.10+ — Custom App YAML + ZFS dataset persistence
• Ubuntu / Debian — apt + docker compose v2, the cleanest deployment path
• Fedora 39+ — dnf docker-ce + SELinux volume labels + firewalld rules
• Proxmox VE LXC — privileged LXC + Intel iGPU passthrough (top pick for home N100 / N305 PVE users)
• Home Assistant OS add-on — add the SkyView repo and one-click install the yunkan / yunkan-openvino add-on
• RK3588 board — a low-power ARM mini box with Rockchip NPU acceleration (Orange Pi 5 / Radxa Rock 5, etc.); 4GB board recommended, 8G+ experimental

Manual deploy (advanced)

If you want to control the process step by step, follow the 4 steps below. The one-line script is essentially these commands automated.

1. 1

   Pull the image

   Pick one of the four variants above:

   bash复制

   docker pull registry.cn-hangzhou.aliyuncs.com/yunkan/yunkan-cpu:latest

2. 2

   Create data / recordings directories

   Small state (database / cookies / logs) lives in data/; recordings (can reach TB scale) go in a sibling recordings/ so you can point them at a large disk / NAS separately.

   bash复制

   mkdir -p ~/skyview/data ~/skyview/recordings && cd ~/skyview

3. 3

   Start the container

   Use host networking so cameras reach the host ports directly, avoiding NAT degrading video quality. License binding requires bind-mounting /etc/machine-id and product_uuid — don't skip them.

   bash复制

   docker run -d --name yunkan --restart=always \
     --network host \
     -v $(pwd)/data:/app/data \
     -v $(pwd)/recordings:/app/data/recordings \
     -v /etc/machine-id:/etc/machine-id:ro \
     -v /sys/class/dmi/id/product_uuid:/sys/class/dmi/id/product_uuid:ro \
     registry.cn-hangzhou.aliyuncs.com/yunkan/yunkan-cpu:latest

4. 4

   Open the browser

   The first visit goes to the /setup wizard; continue with the First Launch chapter.

   复制

   http://<server IP>:23406

Ports and volume mapping

SkyView runs in host network mode, so all ports bind directly on the host. The production ports use an unusual offset into the 2xxxx range to avoid clashing with an existing nginx on 80/443, redis on 6379, or other services on 8080/8554/8888.

Upgrading

Easiest: one-click upgrade from the web console. Open Web Console → Settings → System → Check for updates, and click "Upgrade" when a new version shows up — the system pulls the new image and recreates the container automatically (the database is backed up beforehand, and if the new version fails to start it automatically rolls back). The page reconnects to the new version within a few minutes.

You can also upgrade manually over SSH:

cd ~/skyview
docker compose -f compose.yml pull
docker compose -f compose.yml up -d

Or re-run the one-line script (with -y to skip prompts):

curl -fsSL https://cdn.yun-kan.com/yunkan-install.sh | bash -s -- -y

Uninstalling

cd ~/skyview
docker compose -f compose.yml down
docker rmi registry.cn-hangzhou.aliyuncs.com/yunkan/yunkan-cpu:latest
# Recordings / database are kept; for a full cleanup:
rm -rf ~/skyview