The device(s) work fine in esphome, problem is that for unknown reasons they put audio out on the esp32 and mic + display on the esp32-s3, makes it hard to use it as i voice assistant. so i put it back in the drawer 
you can flash both esp’s from esphome, just flip the usb-c plug.
here is what i got for the esp32-s3 side before i put it back in the drawer:
substitutions:
name: esphome-web-d1e9b4
friendly_name: Xiaozhi Knob
esphome:
name: ${name}
friendly_name: ${friendly_name}
min_version: 2025.5.0
name_add_mac_suffix: false
esp32:
board: esp32-s3-devkitc-1
flash_size: 16MB
cpu_frequency: 240MHz
framework:
type: esp-idf
sdkconfig_options:
CONFIG_ESP32S3_DEFAULT_CPU_FREQ_240: "y"
CONFIG_ESP32S3_DATA_CACHE_64KB: "y"
CONFIG_ESP32S3_DATA_CACHE_LINE_64B: "y"
psram:
mode: octal
speed: 80MHz
api:
ota:
- platform: esphome
id: ota_esphome
logger:
level: DEBUG
wifi:
ssid: !secret wifi_ssid
password: !secret wifi_password
ap:
ssid: "Xiaozhi 3.5 Hotspot"
password: "RZ7D3EzJdPM6"
captive_portal:
uart:
id: audio_uart
tx_pin: GPIO41
rx_pin: GPIO40
baud_rate: 115200
i2c:
- id: bus_a
sda: GPIO11
scl: GPIO12
scan: true
sensor:
- platform: adc
pin: 1
name: "Battery Voltage"
id: battery_voltage
attenuation: 12db
accuracy_decimals: 2
update_interval: 1s
unit_of_measurement: "V"
icon: mdi:battery-medium
filters:
- multiply: 2.0
- median:
window_size: 7
send_every: 7
send_first_at: 7
- throttle: 1min
on_value:
then:
- component.update: battery_percentage
- platform: rotary_encoder
id: knob
name: "Rotary Knob Raw"
pin_a:
number: GPIO8
mode: INPUT_PULLUP
pin_b:
number: GPIO7
mode: INPUT_PULLUP
resolution: 2
on_clockwise:
then:
- logger.log: "Knob turned → CLOCKWISE"
on_anticlockwise:
then:
- logger.log: "Knob turned → COUNTER-CLOCKWISE"
- platform: template
id: battery_percentage
name: "Battery Percentage"
lambda: return id(battery_voltage).state;
accuracy_decimals: 0
unit_of_measurement: "%"
icon: mdi:battery-medium
filters:
- calibrate_linear:
method: exact
datapoints:
- 2.80 -> 0.0
- 3.10 -> 10.0
- 3.30 -> 20.0
- 3.45 -> 30.0
- 3.60 -> 40.0
- 3.70 -> 50.0
- 3.75 -> 60.0
- 3.80 -> 70.0
- 3.90 -> 80.0
- 4.00 -> 90.0
- 4.20 -> 100.0
- lambda: |-
if (x > 100) return 100;
if (x < 0) return 0;
return x;
switch:
- platform: gpio
name: "PA CTRL"
pin: 48
restore_mode: RESTORE_DEFAULT_ON
i2s_audio:
- id: i2s_in
i2s_lrclk_pin: GPIO45
i2s_bclk_pin: GPIO42
microphone:
- platform: i2s_audio
id: i2s_microphone
i2s_audio_id: i2s_in
i2s_din_pin: GPIO46
adc_type: external
pdm: false
channel: left
sample_rate: 16000
bits_per_sample: 32bit
time:
- platform: sntp
id: sntp_time
timezone: "Europe/Berlin"
font:
- file: "gfonts://Roboto"
id: my_font
size: 28
- file: "gfonts://Roboto"
id: roboto_52
size: 52
binary_sensor:
- platform: template
name: "Touch Button"
id: touch_input
# SCREEN & TOUCH
output:
- platform: ledc
pin: 47
id: backlight_output
light:
- platform: monochromatic
id: Sled
name: Screen
icon: "mdi:television"
entity_category: config
output: backlight_output
restore_mode: ALWAYS_ON
default_transition_length: 250ms
external_components:
- source: github://pr#10392
components: [mipi_spi]
refresh: 1d
spi:
id: display_qspi
type: quad
clk_pin: 13
data_pins: [15, 16, 17, 18]
touchscreen:
- platform: cst816
id: my_touchscreen
interrupt_pin: GPIO9
reset_pin: GPIO10
display: main_display
on_touch:
then:
- logger.log:
format: "Touch at (%d, %d)"
args: [touch.x, touch.y]
- binary_sensor.template.publish:
id: touch_input
state: ON
on_release:
then:
- binary_sensor.template.publish:
id: touch_input
state: OFF
display:
- platform: mipi_spi
id: main_display
model: JC3636W518V2
rotation: 180
cs_pin: 14
reset_pin: 21
update_interval: 1s
lambda: |-
auto black = Color(0,0,0), red = Color(255,0,0), green = Color(0,255,0), blue = Color(0,0,255), yellow = Color(255,255,0), white = Color(255,255, 255), cyberlightgreen = Color(0,255,159), cyberlightblue = Color(0,184,255), cyberpink = Color(214,0,255);
it.fill(black);
it.strftime(180, 180, id(roboto_52), cyberlightgreen, TextAlign::CENTER, "%H:%M:%S", id(sntp_time).now());
