SwitchBot AI Hub Review

When SwitchBot announced its AI Hub a few weeks ago, I did not pay much attention to it, assuming it was a typical smart home hub with AI thrown in as a marketing buzzword. We have all seen plenty of products slap ‘AI’ on the box without offering anything meaningful, and I expected this to be more of the same.

However, SwitchBot were kind enough to send me a model to review, and it turns out to be much more than a generic hub. Rather than being just a proprietary smart home hub, this diminutive device also supports Home Assistant and Frigate and can support third-party RTSP-enabled cameras. This makes it quite a unique hub that bridges the gap between proprietary devices and self-hosted solutions, which often require a moderate amount of technical knowledge to set up.

The SwitchBot AI Hub is powered by a local AI chip rated at 6 TOPS (Trillions of Operations Per Second), designed to run a Vision Language Model (VLM) directly on the device. This is what sets it apart from traditional smart home hubs. Rather than relying entirely on cloud processing for AI tasks, much of the object detection and analysis happens on the device itself.

The core specifications are as follows:

• Dimensions: 126 x 94 x 26 mm
• Weight: 235 g
• Internal storage: 32 GB
• RAM: 8 GB
• Storage expansion: microSD up to 1 TB, USB storage or HDD up to 16 TB
• Connections: 1x USB-C 3.0, 1x USB-C 2.0, 1x microSD slot, DC barrel jack
• Network: Wi-Fi 2.4 GHz and 5 GHz, Bluetooth Low Energy
• Camera support: Up to 8 cameras (SwitchBot and RTSP third-party)
• Smart home: Matter bridge for up to 30 devices, control of over 100 SwitchBot devices
• Software: Frigate NVR, Home Assistant Core (containerised), OpenClaw support
• Power: 12 V / 1.5 A barrel jack power supply (included)

The hub runs three containers internally. The first handles Home Assistant, the second runs OpenClaw, and the third runs Frigate NVR, which supports up to eight 2K cameras via both SwitchBot cameras and third-party RTSP feeds. The 8 GB of RAM and 32 GB of internal storage provide a reasonable foundation, though the included 16 GB microSD card will fill up quickly if you are recording multiple camera feeds.

One of the more practical hardware features is the USB-C 3.0 port, which can accept an external SSD for expanded storage up to 16 TB. Alternatively, you can use this port with a USB-C to Ethernet adapter for a wired network connection. SwitchBot states the hub can handle 2.5 Gigabit Ethernet speeds through this method, which is useful if you want the most reliable connection possible. On my 2500 Mbps CityFibre FTTP line, a wired connection would be the preferred option for NVR duties.

The SwitchBot AI Hub is a compact, understated device. The top is finished in matt black plastic, while the underside uses a metal plate that appears to be aluminum, which likely helps with passive heat dissipation. Despite running multiple containers and AI processing tasks, the device does not get noticeably warm during use, and there are no fans or ventilation slots to worry about.

At just 26 mm tall and weighing 235 g, it is small enough to tuck behind a monitor or sit on a shelf without drawing attention. Two rubber feet on the underside keep it stable on flat surfaces. On the rear, you will find the DC barrel jack for power, a reset button, and two USB-C ports. The microSD card slot sits on one side and is slightly recessed, which makes inserting and removing cards a bit fiddly, but it does mean the card sits flush with the housing once installed.

Overall, the design is functional rather than flashy. It does not need to be attractive since it will likely be hidden away in a cabinet or behind other equipment, and the compact form factor means it will not take up much space wherever you put it.

Initial setup is straightforward. The app connects to the hub via Bluetooth first, then walks you through connecting it to your Wi-Fi network. You need to enter your SSID and password, after which the hub connects and appears on your dashboard. On my UniFi network, this process completed without any issues.

After the initial pairing, the hub runs a firmware update. This is automatic and does not require any intervention, but it does take a few minutes. Once the update completes, the hub is ready for camera pairing and further configuration.

From the SwitchBot app, you can access the setup pages for each of the three containers. The app provides the local IP address and port number for Home Assistant, Frigate, and OpenClaw (but this wasn’t available for me at the time of writing), so you can access each service from any device on the same network via a web browser. This is a clean approach that avoids the need to hunt for IP addresses manually, and it means you can manage the hub from a laptop, desktop, or tablet as well as the app itself.

To enable the advanced AI features (without using OpenClaw) you need to subscribe to the AI+ service. This is £6.99 per month, free for the first month and at the time of writing, discounted to £4.99 for subsequent months.

With the subscription, you get the standard advanced object detection, such as human and pet recognition, plus more advanced face detection. However, it then takes things a step further and provides a natural language description of the footage along with the ability to run scenario-driven automations.

So, for example, when I first set up the camera, I was watching TV on the couch, and the camera correctly described me lying on the couch wearing a grey sweater and dark trousers. This is more than simple motion detection or basic person recognition. The VLM analyses the scene and produces a contextual summary of what it sees, which can then be used as a trigger for automations.

In testing, the AI descriptions were impressively detailed. The system also categorizes events into useful groups, such as Security Manager, Care Manager, and Pet Manager, making it easier to filter relevant notifications.

The free tier still provides local AI detection for humans, pets, vehicles, and general motion, which is functional for basic use. The subscription adds the VLM-powered contextual analysis, daily event summaries, and the ability to search through footage using natural language queries. Whether the subscription is worth it depends on how much you value those advanced features, but the base functionality without it is still more capable than most competing hubs.

With the AI Hub set up, you can begin adding cameras. SwitchBot sent me a 3K Pan/Tilt Cam Plus to test alongside the hub. You need to add the camera to the SwitchBot app first and make sure it is running a compatible firmware, and you can then add it to the hub. The SwitchBot camera pairing process involves scanning a QR code displayed in the app with the camera itself, and it connects within seconds. The pan and tilt motors are whisper-quiet, and the camera controls are responsive with minimal lag.

For third-party cameras, I used one of my Reolink models. For this, you need the RTSP feed URL, ideally defining the primary and sub feeds. For Reolink, the URLs look like:

• Main Stream: rtsp://admin:111111@192.168.10.92/Preview_01_main
• Sub Stream: rtsp://admin:111111@192.168.10.92/Preview_01_sub

The SwitchBot camera works perfectly with the hub, as you would expect. However, I have found the Reolink camera to be a bit temperamental. It appears to struggle to load the primary feed and then encounters artifacts on the stream. The motion detection appears to work without issue on the Reolink, and I suspect the main problem is with the Reolink cameras themselves rather than the hub. Other reviewers have reported similar RTSP compatibility issues with certain camera brands, and SwitchBot is actively working on improving third-party camera support through firmware updates.

The SwitchBot AI Hub acts as a Matter bridge, exposing up to 30 SwitchBot sub-devices to ecosystems like Apple Home, Google Home, Amazon Alexa, and SmartThings. This is not a new feature for SwitchBot hubs, as the Hub Mini Matter and Hub 3 already provide this functionality, but it is expected at this price point.

The 30-device Matter bridge limit could be restrictive for larger smart home setups. If you have a substantial number of SwitchBot devices spread across multiple rooms, you may find yourself bumping up against this ceiling. That said, the hub can control over 100 SwitchBot devices directly through Bluetooth and Wi-Fi, so the Matter bridge limit only affects cross-ecosystem exposure.

One limitation at the time of writing is that SwitchBot cameras are not yet exposed to Apple Home via the hub. This will require Matter 1.5 support from both the hub and Apple Home, which is not yet available. For now, camera management and viewing are handled entirely through the SwitchBot app, Frigate, or Home Assistant.

One pleasant surprise was the ability to install Frigate, the open-source NVR. This may seem redundant given that the SwitchBot app can handle events and notifications on its own, but the Frigate installation provides several advantages. It allows you to access the camera streams via a web browser from any device on the network, and it supports a wide variety of advanced AI object detection options that offer much more customization than the SwitchBot app alone.

Frigate on the AI Hub appears to be a full version of the software. You get the live view of all connected cameras, a review tab with a timeline that lets you scrub through 24 hours of footage, and face detection training.

The event detection in Frigate is detailed. It recognizes objects like couches, cups, TVs, and people, and provides hover-over previews of events on the timeline. For anyone familiar with Frigate on a dedicated server, the experience is essentially the same, just running on this compact hub rather than a mini PC or Raspberry Pi.

Frigate also experienced issues with my Reolink camera, which is probably more evidence to suggest the issue is with the camera rather than the hub. One thing to be aware of is that Frigate playback can appear slightly stuttery. This is a known characteristic of Frigate in general and not specific to the SwitchBot AI Hub. For smooth continuous recording, you would ideally pair Frigate with a separate NVR for footage storage while using Frigate for AI detection and event management. At the time of testing, the system was running at approximately 25% CPU and 4% GPU utilization, suggesting there is headroom for additional cameras.

The Home Assistant installation on the AI Hub is a containerized version of Home Assistant Core. This means it runs as a Docker container managed by SwitchBot, which has both advantages and disadvantages.

On the positive side, it is pre-installed and ready to go. You access it by typing the hub’s IP address with port 8123 into a browser, and you are presented with the standard Home Assistant setup wizard. The system automatically discovered most of my Wi-Fi smart home devices, including my Philips Hue bulbs. I was able to create a basic dashboard and control devices within minutes. The integration