The SuperSensor is an all-in-one voice, motion, presence, temperature/humidity/air quality, and light sensor, built on an ESP32 with ESPHome, and inspired heavily by the EverythingSmartHome Everything Presence One sensor and the HomeAssistant "$13 Voice Assistant" project.

Use SuperSensors around your house to provide HomeAssistant Voice Assist interfaces with wake word detection, as well as other sensor detection options as you want them.

Assist feedback is provided by a pair of common-cathode RGB LED. No speakers or annoying TTS feedback here! With an optional 3D Printed case and a clear diffuser cover, the LEDs can be turned into a sleek light bar on the bottom of the unit for quick and easy confirmation of voice actions, or just use it bare if you like the "PCB on a wall" aesthetic.

To Use:

Install the ESPHome configuration supersensor.yaml to a compatible ESP32 devkit (below).
Install the ESP32 and sensors into the custom PCB.
Power up the SuperSensor, connect to the WiFi AP, and connect it to your network.
Install the SuperSensor somewhere that makes sense.
Add/adopt the SuperSensor to HomeAssistant using the automatic name.
Tune the SuperSensor values to your needs.

Note: Once programmed, the output LED will flash continuously until connected to HomeAssistant, and a bit longer to establish if the wake word functionality is enabled. This is by design, so you know if your sensors are connected or not. If you do not want this, comment out the light.turn_on block starting on line 38 of the ESPHome configuration to disable this functionality.

For more details, please see my first blog post on the SuperSensor project and my update post on version 2.0.

NOTE: For those with v1.x hardware, see the repository for that code instead.

Major Changes from 1.x

Replaced the Bosch BME680 with the Sensirion SHT45 and Sensirion SGP41.

The BME680 proved to be woefully unreliable in my testing. Temperature was fairly accurate (internal heating and offset notwithstanding), but humidity was wildly off of what other thermometers/hydrometers would report. In addition, the AQ functionality of the sensor was a source of much frustration and I was never able to get it to work reliably, either with the official BSEC library or with my own attempts at self-configuration.

Thus, this sensor has been replaced with two Sensirion sensors which in my experience so far have been much more reliable and consistent. There is a slight cost increase due to these sensors, but not signfigant enough to outweigh the benefit of reliable monitoring they confer.
Replaced the SR602 PIR sensor with the AM312 PIR sensor.

The SR602 was, in my experience, prone to constant false misfirings even in completely empty rooms. In addition its orientation requirements are awkward (pins on the left or right side). While it's possible I just had a bad batch, this soured me significantly to these sensors, especially after reading many other similar reports around the internet.

Thus, this sensor has been replaced with the more reliable AM312. While the form factor fo the AM312 leaves a bit to be desired (sticking up by about 1cm more from the board), in the end I found this to be more of a positive than a negative for resposiveness and apperance.
Completely redesigned the custom PCB around the above sensor changes, which is now more compact in a 50x55mm almost-square configuration.
Significantly cleaned up the ESPHome configuration, to support the above sensors and remove a lot of cruft that was caused by the BME680.

Parts List

Qty	Component	Cost (2025/05 CAD, ex. shipping)	Links
1	GY-SGP41	$11.08	AliExpress
1	GY-SHT45	$5.67	AliExpress*
1	SR602	$0.81	AliExpress
1	TSL2591	$4.59	AliExpress
1	HL-LD2510C	$4.79	AliExpress*
1	INMP441	$2.93	AliExpress
1	ESP32 HW-395	$6.67	AliExpress*
2	RBG LED	$0.09 ($9.12/100)	Amazon
1	470Ω resistor	$0.08 ($7.99/100)	Amazon
2	Female pin header†	$1.59 ($15.99/10)	Amazon
1	Custom PCB (JLC)	$0.69 ($6.89/10)	GitHub
TOTAL		$40.58

* Ensure you select the correct device on the page as it shows multiple options.

† One of these sets is optional, and is useful if you do not want to solder the individual sensors directly to the board (see below).

To Solder or Not To Solder

Personally, for my Supersensor 1.x's and the initial batch of Supersensor 2.x's, I directly soldered all the non-ESP components to the board. This proved to be a major mistake when I later decided to switch from SGP30's to SGP41's after some testing and I had to desolder all of them, ruining several PCBs in the process. It was also a hassle to desolder the existing sensors for reuse during the 1.x to 2.x conversion.

As a result, I actually strongly encourage anyone building one of these units to leverage sockets for all components, to allow for quick swapping if any turn out to be defective or if future changes are warranted.

Note that due to the PCB design, you must socket at least one set of components - either the ESP32 or the sensors on the front. Due to the positioning and overlap, it would be impossible to solder everything directly to the board, as the ESP covers several of the solder points of the front sensors and vice versa.

You can use the provided 40-pin female headers exclusively if you wish, and cut them to length for the individual sensors as needed, or you can use individually-sized female headers in the following quantities should you wish for a slightly neater finish:

3x 3-pin (AM302, INMP441 x2)
2x 4-pin (SGP41, SHT45)
1x 5-pin (LD2410C)
1x 6-pin (TSL2591)

I will leave it up to the reader to source these specific sizes if they desire (I found all except a 5-pin on Amazon, and just used a 6-pin with one pin removed).

I still directly solder the RGB LEDs and resistor to the board for simplicity as these very small leads are not easily socketed, and these components are so inexpensive as to be effectively disposable along with the PCB should that be required.

Part Swaps

To save a little money, it is possible to swap out the two Sensirion sensors for their less-feature- rich peers, with no code changes:

SGP41 -> SGP40 - removes the NOx functionality
SHT45 -> SHT40/41/43 - less accuracy

Personally, I do not find the minimal cost savings to be worth sacrificing the extra potential functionality, so I recommend using the provided models, but this is up to the builder to decide.

No other parts can be easily swapped without code or PCB design changes.

Configurable Options

There are several UI-configurable options with the SuperSensor to help you get the most out of the sensor for your particular use-case.

Note: Configuration of the LD2410C is excluded here, as it is extensively configurable. See the documentation for more details on its options.

Enable Voice Support (switch)

If enabled (the default), the SuperSensor's voice functionality including wake word will be started. Disabling this defeats most of the point of the SuperSensor, but can be done if desired, for instance if you have multiple SuperSensors in a single room and only want one to respond to voice commands.

Enable Presence LED (switch)

If enabled (the default), when overall presence is detected, the LEDs will glow "white" at 15% power to signal presence.

Temperature Offset (selector, -30 to +10 @ 0.1, -5 default)

Allows calibration of the SHT45 temperature sensor with an offset from -30 to +10 degrees C. Useful if the sensor is misreporting actual ambient tempreatures. Due to internal heating of the SHT45 by the ESP32, this defaults to -5; further calibration may be needed for your sensors and environment based on an external reference.

Humidity Offset (selector, -20 to +20 @ 0.1)

Allows calibration of the SHT45 humidity sensor with an offset from -10 to +10 percent relative humidity. Useful if the sensor is misreporting actual humidity based on an external reference.

PIR Hold Time (selector, 0 to +60 @ 5, 0 default)

The SuperSensor uses an AM312 PIR sensor, which has a stock hold time of ~2.5 seconds. This setting allows increasing that value, with retrigger support, to up to 60 seconds, allowing the PIR detection to report for longer. 0 represents "as long as the AM312 fires".

Light Threshold Control (selector, 0 to +200 @ 5, 30 default)

The SuperSensor features a "light presence" binary sensor based on the light level reported by the TSL2591 sensor. This control defines the minimum lux value from the sensor to be considered "presence". For instance, if you have a room that is usually dark at 0-5 lux, but illuminated to 100 lux when a (non-automated) light switch is turned on, you could set a threshold here of say 30 lux: then, while the light is on, "light presence" is detected, and when the light is off, "light presence" is cleared. Light presence can be used standalone or as part of the integrated occupancy sensor (below).

Integrated Occupancy Sensor (Selector)

The SuperSensor features a fully integrated "occupancy" sensor, which can be configured to provide exactly the sort of occupancy detection you may want for your room.

There are 7 options (plus "None"/disabled), with both "detect" and "clear" handled separately:

PIR + Radar + Light

Occupancy is detected when all 3 sensors report detected, and occupancy is cleared when any of the sensors report cleared.

For detect, this provides the most "safety" against misfires, but requires a normally-dark room with a non-automated light source and clear PIR detection positioning.

For clear, this option is probably not very useful as it is likely to clear quite frequently from the PIR, but is provided for completeness.

PIR + Radar

Occupancy is detected when both sensors report detected, and occupancy is cleared when either of the sensors report cleared.

For detect, this provides good "safety" against PIR misfires without needing a normally-dark room, though detection may be slightly delayed from either sensor.

For clear, this option is probably not very useful as it is likely to clear quite frequently from the PIR, but is provided for completeness.

PIR + Light

Occupancy is detected when both sensors report detected, and occupancy is cleared when either of the sensors report cleared.

For detect, this provides some "safety" against PIR misfires, but requires a normally-dark room with a non-automated light source and clear PIR detection positioning.

For clear, this option is probably not very useful as it is likely to clear quite frequently from the PIR, but is provided for completeness.

Radar + Light

Occupancy is detected when both sensors report detected, and occupancy is cleared when either of the sensors report cleared.

For detect, this allows for radar detection while suppressing occupancy without light, for instance in a hallway where one might not want a late night bathroom visit to turn on the lights, or something to that effect.

For clear, this option can provide a useful option to clear presence quickly if the lights go out, while still providing Radar presence.

PIR Only

Occupancy is based entirely on the PIR sensor for both detect and clear.

Prone to misfires, but otherwise a good option for quick detection and clearance in a primarily-moving zone (e.g. hallway).

Radar Only

Occupancy is based entirely on the Radar sensor for both detect and clear.

Useful for an area with no consistent motion or light level.

Light Only

Occupancy is based entirely on the Light sensor for both detect and clear.

Useful for full dependence on an external light source.

None

Disable the functionality in either direction.

For detect, no occupancy will ever fire.

For clear, no states will clear occupancy; with any detect option, this means that occupancy will be detected only once and never clear, which is likely not useful.

AQ Details

The SuperSensor 2.0 features an SGP41 air quality sensor by Sensirion. This is a powerful AQ sensor which powers several commercial devices including the AirGradient One, which gave us a lot of our configuration via their sharing of algorithms.

The sensor provides two base readings: a VOC Index, and a NOx Index. These values are both floating references centered at 100 (VOC) and 1 (NOx), where that value represents "normal" air over the previous 24 hours. These sensors are very useful for any sort of quick-change automations, e.g. turn on a fan if levels spike due to cooking.

In addition, we leverage AirGradient's published forumulas to convert the VOC index into actual VOC quantities, in both µg/m³ and ppb. While this may drift due to the sensor's regular internal recalibration, I feel that following what AirGradient does is sufficient enough for any real-world home usage. Further, we use a very rough conversion of the aforementioned VOC quantity into an eCO2 reading, using Isobutylene as a reference gas. These sensors are more useful for display purposes, to show the current levels in a room in a dashboard or other such place, for human consumption. Note that no such conversions are done for NOx as there are no (that I can find) published empirical calculations for this conversion, unlike for VOCs via AirGradient.

Note however that like all MOx sensors, the SGP41 does not differentiate gasses, and as such cannot tell the difference between normal, everyday natural VOCs like those in breath or from e.g. ripening fruit, and dangerous VOCs from e.g. construction materials. These should be used only as a general indication of air quality over short periods, rather than an absolute reference over long periods (much to my own frustration but inevitable begruding acceptance).