diff --git a/Makefile b/Makefile
new file mode 100644
index 0000000..4c7662d
--- /dev/null
+++ b/Makefile
@@ -0,0 +1,63 @@
+MCU_TARGET     = atmega328p
+MCU_FREQ       = 16000000UL
+MCU_TTY        = /dev/ttyUSB0
+MCU_TTY_BAUD   = 57600
+MCU_PROGRAMMER = arduino
+
+PRG            = main
+OBJ            = main.o uart.o
+
+OPTIMIZE       = -O2
+DEFS           = -DF_CPU=$(MCU_FREQ) -DBAUDRATE=$(MCU_TTY_BAUD)
+LIBS           =
+# You should not have to change anything below here.
+CC             = avr-gcc
+# Override is only needed by avr-lib build system.
+override CFLAGS        = -g --std=c99 -Wall $(OPTIMIZE) -mmcu=$(MCU_TARGET) $(DEFS)
+override LDFLAGS       = -Wl,-Map,$(PRG).map
+OBJCOPY        = avr-objcopy
+OBJDUMP        = avr-objdump
+
+all: $(PRG).elf $(PRG).asm
+
+$(PRG).elf: $(OBJ)
+	$(CC) $(CFLAGS) $(LDFLAGS) -o $@ $^ $(LIBS)
+
+# dependencies:
+
+clean:
+	rm -rf *.o $(PRG).elf *.eps *.png *.pdf *.bak *.asm *.hex
+	rm -rf *.lst *.map $(EXTRA_CLEAN_FILES)
+
+lst:  $(PRG).lst
+
+%.lst: %.elf
+	$(OBJDUMP) -h -S $< > $@
+
+# Rules for building the .text rom images
+text: hex bin srec
+
+hex:  $(PRG).hex
+
+bin:  $(PRG).bin
+
+%.hex: %.elf
+	$(OBJCOPY) -j .text -j .data -O ihex $< $@
+
+%.bin: %.elf
+	$(OBJCOPY) -j .text -j .data -O binary $< $@
+
+# to get the final asm code
+
+%.asm: %.elf
+	$(OBJDUMP) -z -j .text -m avr5 -d $< > $@
+
+reset:
+	avrdude -v -p$(MCU_TARGET) -c$(MCU_PROGRAMMER) -P$(MCU_TTY) -b$(MCU_TTY_BAUD)
+
+flash: $(PRG).hex
+	avrdude -v -p$(MCU_TARGET) -c$(MCU_PROGRAMMER) -P$(MCU_TTY) -b$(MCU_TTY_BAUD) -u -U flash:w:$(PRG).hex
+
+screen:
+	screen $(MCU_TTY) $(MCU_TTY_BAUD)
+
diff --git a/main.c b/main.c
new file mode 100644
index 0000000..6720334
--- /dev/null
+++ b/main.c
@@ -0,0 +1,114 @@
+#include <stdio.h>
+#include <avr/io.h>
+#include <util/delay.h>
+#include <avr/interrupt.h>
+#include "uart.h"
+
+/* Button matrix between PB3-PB4, PC0-PC3
+ * PC0-PC3 are inputs, default pull-up
+ * PB3-PB4 are pulled to ground (alternating) while the above are read
+ */
+
+void print_nibble(uint8_t nibble) {
+        nibble += 0x30;
+        if (nibble > 0x39) nibble += 7;
+        uart_putc(nibble);
+}
+
+void print_int8hex(uint8_t val) {
+        print_nibble(val >> 4);
+        print_nibble(val & 0x0F);
+}
+
+uint8_t getButtonState() {
+        uint8_t val = 0;
+
+        // Pull upper button line (PB4) down
+        PORTB |= (1 << PB3);
+        PORTB &= ~(1 << PB4);
+	_delay_ms(1);
+	val = ((~PINC) << 4);
+
+        // Pull lower button line (PB3) down
+        PORTB &= ~(1 << PB3);
+        PORTB |= (1 << PB4);
+	_delay_ms(1);
+	val |= (~PINC) & 0x0F;
+
+	return val;
+}
+
+uint8_t getrelay(uint8_t n) {
+	uint8_t v = 0;
+	switch(n) {
+		case 0: v = (PIND >> PD2); break;
+		case 1: v = (PIND >> PD3); break;
+		case 2: v = (PIND >> PD4); break;
+		case 3: v = (PIND >> PD5); break;
+		case 4: v = (PIND >> PD6); break;
+		case 5: v = (PIND >> PD7); break;
+		case 6: v = (PINB >> PB0); break;
+		case 7: v = (PINB >> PB1); break;
+	}
+	return(~v & 1);
+}
+
+void setrelay(uint8_t n, uint8_t s) {
+	if (s) {
+		switch(n) {
+                        case 0: PORTD &= ~(1 << PD2); break;
+                        case 1: PORTD &= ~(1 << PD3); break;
+                        case 2: PORTD &= ~(1 << PD4); break;
+                        case 3: PORTD &= ~(1 << PD5); break;
+                        case 4: PORTD &= ~(1 << PD6); break;
+                        case 5: PORTD &= ~(1 << PD7); break;
+                        case 6: PORTB &= ~(1 << PB0); break;
+                        case 7: PORTB &= ~(1 << PB1); break;
+		}
+	} else {
+		switch(n) {
+                        case 0: PORTD |= (1 << PD2); break;
+                        case 1: PORTD |= (1 << PD3); break;
+                        case 2: PORTD |= (1 << PD4); break;
+                        case 3: PORTD |= (1 << PD5); break;
+                        case 4: PORTD |= (1 << PD6); break;
+                        case 5: PORTD |= (1 << PD7); break;
+                        case 6: PORTB |= (1 << PB0); break;
+                        case 7: PORTB |= (1 << PB1); break;
+		}
+	}
+}
+
+int main() {
+	DDRC &= ~(0b1111); // This sets PC0 to PC3 to input
+	PORTC |= 0b1111; // This makes dito pull-up per default
+
+	DDRB |= (1 << PB3) | (1 << PB4); // Keybad rows
+
+	DDRD |= 0b11111100; // PD2 to PD7 for relay
+	DDRB |= 0b00000011; // PB0 and PB1 for relay
+
+	// Same with PORT, so relays stay off
+	PORTD |= 0b11111100;
+	PORTB |= 0b00000011;
+
+	sei();
+	uint8_t c,p,d,i;
+	p = 0;
+
+	while (1) {
+		c = getButtonState();
+		d = c ^ p;
+		p = c;
+		for (i = 0; i<8; i++) {
+			if (d & c & 0x80) {
+				setrelay(i, 1 - getrelay(i));
+				print_nibble(i);
+				uart_putc('\r');
+				uart_putc('\n');
+			}
+			d = d << 1;
+			c = c << 1;
+		}
+	};
+}
diff --git a/uart.c b/uart.c
new file mode 100644
index 0000000..ddf511d
--- /dev/null
+++ b/uart.c
@@ -0,0 +1,82 @@
+#include <avr/io.h>
+#include <avr/interrupt.h>
+#include <util/atomic.h>
+
+#include "uart.h"
+
+#define UART_BUFSIZE 64
+
+typedef volatile struct {
+	char data[UART_BUFSIZE];
+	uint8_t r;
+	uint8_t w;
+} buf;
+
+buf __attribute__ ((section (".noinit"))) uart_rxbuf;
+buf __attribute__ ((section (".noinit"))) uart_txbuf;
+
+#define BAUD_PRESCALE (((F_CPU/(BAUDRATE*8UL)))-1)
+
+void __attribute__ ((naked)) __attribute__ ((section (".init5"))) uart_init() {
+	uart_rxbuf.r = 0;
+	uart_rxbuf.w = 0;
+
+	uart_txbuf.r = 0;
+	uart_txbuf.w = 0;
+
+	UCSR0A = (1<<U2X0);
+	UCSR0B = (1<<RXEN0) | (1<<TXEN0) | (1<<RXCIE0);
+	UCSR0C = (1<<UCSZ00) | (1<<UCSZ01); // 8N1
+	UBRR0 = BAUD_PRESCALE;
+}
+
+void __attribute__ ((naked)) __attribute__ ((section (".fini5"))) uart_deinit() {
+	while ((UCSR0B & (1 << UDRIE0))) {};
+	while (!(UCSR0A & (1 << TXC0))) {};
+}
+
+// RX Facilities
+ISR(USART_RX_vect) {
+	uint8_t c = UDR0;
+
+	if (((uart_rxbuf.w - uart_rxbuf.r) & (UART_BUFSIZE - 1)) == 1) return;
+	uart_rxbuf.data[uart_rxbuf.w++] = c;
+	uart_rxbuf.w = uart_rxbuf.w & (UART_BUFSIZE - 1);
+}
+
+uint8_t uart_getc() {
+	uint8_t c;
+	while (uart_rxbuf.w == uart_rxbuf.r) {};
+
+	// Disable interrupt to avoid race condition
+	ATOMIC_BLOCK(ATOMIC_RESTORESTATE) {
+		c = uart_rxbuf.data[uart_rxbuf.r++];
+		uart_rxbuf.r = uart_rxbuf.r & (UART_BUFSIZE - 1);
+	}
+
+	return c;
+}
+
+// TX Facilities
+ISR(USART_UDRE_vect) {
+	if ((uart_txbuf.w != uart_txbuf.r) && (UCSR0B & (1<<TXEN0))) {
+		UDR0 = uart_txbuf.data[uart_txbuf.r++];
+		uart_txbuf.r = uart_txbuf.r & (UART_BUFSIZE - 1);
+	} else {
+		UCSR0B &= ~(1<<UDRIE0); // UDRE off
+	}
+}
+
+uint8_t uart_putc(unsigned char c) {
+	while (((uart_txbuf.r - uart_txbuf.w) & (UART_BUFSIZE - 1)) == 1) {};
+
+	ATOMIC_BLOCK(ATOMIC_RESTORESTATE) {
+		uart_txbuf.data[uart_txbuf.w++] = c;
+		uart_txbuf.w = uart_txbuf.w & (UART_BUFSIZE - 1);
+	}
+
+	UCSR0B |= (1<<UDRIE0); // UDRE on
+
+	return 1;
+}
+
diff --git a/uart.h b/uart.h
new file mode 100644
index 0000000..ead435b
--- /dev/null
+++ b/uart.h
@@ -0,0 +1,4 @@
+#include <stdint.h>
+
+uint8_t uart_getc();
+uint8_t uart_putc(unsigned char c);