Microphone - INMP441 (I2S)

SD    -> GPIO 40
WS    -> GPIO 41
SCK   -> GPIO 42
L/R   -> GND
VDD   -> 3.3V
GND   -> GND

Amplifier - MAX98357A
DIN   -> GPIO 17   (shared I2S data)
BCLK  -> GPIO 47   (shared)
LRC   -> GPIO 21   (shared)
SD    -> 3.3V, or GPIO 38 if you fit the PCM5102A
VIN   -> 5V
GAIN  -> 3.3V      (6 dB)
GND   -> GND

Headphone DAC - PCM5102A (optional)
DIN   -> GPIO 17   (shared)
BCK   -> GPIO 47   (shared)
LCK   -> GPIO 21   (shared)
SCK   -> GND       (internal PLL mode)
XSMT  -> GPIO 39   (HIGH = play)
FMT / FLT / DEMP -> GND
VIN   -> 3.3V
GND   -> GND

Display - GC9A01, 240x240 round SPI
SCLK  -> GPIO 6
MOSI  -> GPIO 7
CS    -> GPIO 5
DC    -> GPIO 4
RST   -> GPIO 2
BLK   -> 3.3V
VCC   -> 3.3V
GND   -> GND

Button and LED
PTT button -> GPIO 1, INPUT_PULLUP, active LOW
NeoPixel   -> GPIO 48 (built in on most S3 boards)

Power: the 3.3V rail feeds the mic, the display and the optional DAC.
The MAX98357A wants 5V. Common ground throughout.

ESP32S3 Dev Module - CPU 240 MHz - Flash 16 MB - PSRAM: OPI PSRAM - Upload 921600 - Partition scheme: custom, uses partitions.csv.

Libraries: ArduinoJson, Adafruit NeoPixel, Arduino_GFX_Library.

Do NOT install ArduinoWebsockets from the Library Manager. A patched copy is vendored in src/, because 0.5.3 cannot open a WSS connection on ESP32 and fails silently. That one cost me a weekend.

PlatformIO: everything is pinned in platformio.ini.   pio run -t upload   pio run -t uploadfs

The sketch and the emoji frames live in different partitions. Flash the sketch, then upload the LittleFS image separately - 56 pre-rendered RGB565 .bin files, about 11 MB. Serial prints "Display: loaded 8 frames for 'neutral'" when it worked. Skipping this is the usual way to end up staring at a blank screen.

No WiFi or OPENAI credentials are compiled in. On first boot the device opens an open AP called VOICE-AGENT-XXYY, flashes the LED yellow, and puts a QR code on the display. Join it, the captive portal opens by itself, enter your network, Save & Restart. After that it lives at voice-agent-XXYY.local. Credentials persist in NVS.

In the portal, you also need to add your OPENAI key. (Get one here; you get $5 when you register: https://platform.openai.com/ )

• Yellow solid     connecting to WiFi 
• Yellow flashing  AP config mode, waiting for credentials 
• Green            ready, push to talk 
• Pink             recording 
• Blue             waiting on the API 
• Cyan             speaking

Toolchain versions, config.h, the portal admin zone and the parts list are in the repo, which stays current: https://github.com/iollama/Voice-Assistant-5/