diff --git a/.vscode/tasks.json b/.vscode/tasks.json
index 0040b88..4257891 100644
--- a/.vscode/tasks.json
+++ b/.vscode/tasks.json
@@ -9,9 +9,11 @@
                 "-g",
                 "${fileDirname}/main.cpp",
                 "${fileDirname}/Xenith/core.cpp",
+                "${fileDirname}/Xenith/token/token.cpp", 
                 "-o",
                 "${fileDirname}/main",
-                "-I", "${fileDirname}/Xenith"
+                "-I", "${fileDirname}/Xenith",
+                "-I", "${fileDirname}/Xenith/token" 
             ],
             "options": {
                 "cwd": "${fileDirname}"
diff --git a/Xenith/core.cpp b/Xenith/core.cpp
index 29203c0..ed1837e 100644
--- a/Xenith/core.cpp
+++ b/Xenith/core.cpp
@@ -1,82 +1,63 @@
 #include "core.h"
+#include <cmath>
+#include <cstdlib>
 
-NeuralNetwork::NeuralNetwork(LayerStructure_t layers[], int count) {
-    numLayers = count;
-    for (int i = 0; i < count; i++) {
-        layerSizes.push_back(layers[i].size);
-    }
-
-    // Инициализация весов случайными числами
+NeuralNetwork::NeuralNetwork(LayerStructure_t layers[], int count) : numLayers(count) {
+    for (int i = 0; i < count; i++) sizes.push_back(layers[i].size);
     for (int i = 0; i < count - 1; i++) {
-        std::vector<std::vector<double>> layerWeights;
-        for (int j = 0; j < layerSizes[i+1]; j++) {
-            std::vector<double> nodeWeights;
-            for (int k = 0; k < layerSizes[i]; k++) {
-                nodeWeights.push_back(((double)rand() / RAND_MAX) * 2 - 1);
-            }
-            layerWeights.push_back(nodeWeights);
+        std::vector<std::vector<double>> layerW;
+        double scale = sqrt(2.0 / sizes[i]);
+        for (int j = 0; j < sizes[i+1]; j++) {
+            std::vector<double> nodeW;
+            for (int k = 0; k < sizes[i]; k++) 
+                nodeW.push_back(((double)rand()/RAND_MAX * 2 - 1) * scale);
+            layerW.push_back(nodeW);
         }
-        weights.push_back(layerWeights);
-
-        std::vector<double> layerBiases;
-        for (int j = 0; j < layerSizes[i+1]; j++) {
-            layerBiases.push_back(((double)rand() / RAND_MAX) * 2 - 1);
-        }
-        biases.push_back(layerBiases);
+        weights.push_back(layerW);
+        biases.push_back(std::vector<double>(sizes[i+1], 0.0));
     }
 }
 
-std::vector<double> NeuralNetwork::feedForward(std::vector<double> input) {
+std::vector<double> NeuralNetwork::feedForward(const std::vector<double>& input) {
     outputs.clear();
     outputs.push_back(input);
-
-    std::vector<double> current = input;
+    std::vector<double> curr = input;
     for (int i = 0; i < numLayers - 1; i++) {
         std::vector<double> next;
-        for (int j = 0; j < layerSizes[i+1]; j++) {
+        for (int j = 0; j < sizes[i+1]; j++) {
             double sum = biases[i][j];
-            for (int k = 0; k < layerSizes[i]; k++) {
-                sum += current[k] * weights[i][j][k];
-            }
-            next.push_back(sigmoid(sum));
+            for (int k = 0; k < (int)curr.size(); k++) sum += curr[k] * weights[i][j][k];
+            next.push_back(1.0 / (1.0 + exp(-sum)));
         }
-        current = next;
-        outputs.push_back(current);
+        curr = next;
+        outputs.push_back(curr);
     }
-    return current;
+    return curr;
 }
 
-void NeuralNetwork::train(std::vector<double> input, std::vector<double> target, double lr) {
-    // 1. Прямой проход
-    feedForward(input);
-
-    // 2. Вычисление ошибок для выходного слоя
+double NeuralNetwork::train(const std::vector<double>& input, const std::vector<double>& target, double lr) {
+    std::vector<double> pred = feedForward(input);
     std::vector<std::vector<double>> errors(numLayers);
-    errors[numLayers - 1].resize(layerSizes[numLayers - 1]);
-    for (int i = 0; i < layerSizes[numLayers - 1]; i++) {
-        double output = outputs[numLayers - 1][i];
-        errors[numLayers - 1][i] = (target[i] - output) * sigmoidDerivative(output);
+    errors[numLayers-1].resize(sizes[numLayers-1]);
+    double totalErr = 0;
+    for (int i = 0; i < sizes[numLayers-1]; i++) {
+        double e = target[i] - pred[i];
+        errors[numLayers-1][i] = e * pred[i] * (1.0 - pred[i]);
+        totalErr += e * e;
     }
-
-    // 3. Обратное распространение ошибки на скрытые слои
     for (int i = numLayers - 2; i > 0; i--) {
-        errors[i].resize(layerSizes[i]);
-        for (int j = 0; j < layerSizes[i]; j++) {
-            double error = 0.0;
-            for (int k = 0; k < layerSizes[i+1]; k++) {
-                error += errors[i+1][k] * weights[i][k][j];
-            }
-            errors[i][j] = error * sigmoidDerivative(outputs[i][j]);
+        errors[i].resize(sizes[i]);
+        for (int j = 0; j < sizes[i]; j++) {
+            double e = 0;
+            for (int k = 0; k < sizes[i+1]; k++) e += errors[i+1][k] * weights[i][k][j];
+            errors[i][j] = e * outputs[i][j] * (1.0 - outputs[i][j]);
         }
     }
-
-    // 4. Обновление весов и смещений
     for (int i = 0; i < numLayers - 1; i++) {
-        for (int j = 0; j < layerSizes[i+1]; j++) {
-            for (int k = 0; k < layerSizes[i]; k++) {
-                weights[i][j][k] += lr * errors[i+1][j] * outputs[i][k];
-            }
+        for (int j = 0; j < sizes[i+1]; j++) {
+            for (int k = 0; k < sizes[i]; k++) weights[i][j][k] += lr * errors[i+1][j] * outputs[i][k];
             biases[i][j] += lr * errors[i+1][j];
         }
     }
+    return totalErr;
 }
\ No newline at end of file
diff --git a/Xenith/core.h b/Xenith/core.h
index 038f768..dc79cd6 100644
--- a/Xenith/core.h
+++ b/Xenith/core.h
@@ -4,25 +4,22 @@
 #include "typedef.h"
 #include <vector>
 #include <cmath>
-#include <iostream>
-#include <cstdlib>
 
 class NeuralNetwork {
 private:
     int numLayers;
-    std::vector<int> layerSizes;
-    std::vector<std::vector<std::vector<double>>> weights; // weights[layer][to_node][from_node]
-    std::vector<std::vector<double>> biases;             // biases[layer][node]
-    std::vector<std::vector<double>> outputs;            // Храним выходы слоев для backprop
+    std::vector<int> sizes;
+    std::vector<std::vector<std::vector<double>>> weights;
+    std::vector<std::vector<double>> biases;
+    std::vector<std::vector<double>> outputs;
 
     double sigmoid(double x) { return 1.0 / (1.0 + exp(-x)); }
-    double sigmoidDerivative(double x) { return x * (1.0 - x); }
+    double sigmoidDeriv(double x) { return x * (1.0 - x); }
 
 public:
     NeuralNetwork(LayerStructure_t layers[], int count);
-    
-    std::vector<double> feedForward(std::vector<double> input);
-    void train(std::vector<double> input, std::vector<double> target, double learningRate);
+    std::vector<double> feedForward(const std::vector<double>& input);
+    double train(const std::vector<double>& input, const std::vector<double>& target, double lr);
 };
 
 #endif
\ No newline at end of file
diff --git a/Xenith/token/token.cpp b/Xenith/token/token.cpp
new file mode 100644
index 0000000..fa24bd7
--- /dev/null
+++ b/Xenith/token/token.cpp
@@ -0,0 +1,46 @@
+#include "token.h"
+#include <algorithm>
+#include <random>
+
+void Tokenizer::add(std::string word) {
+    int id = wordToId.size();
+    wordToId[word] = id;
+    idToWord[id] = word;
+}
+
+std::string Tokenizer::getWord(int id) {
+    return idToWord.count(id) ? idToWord[id] : "";
+}
+
+std::vector<int> Tokenizer::textToTokens(const std::string& text) {
+    std::vector<int> tokens;
+    size_t pos = 0;
+    while (pos < text.length()) {
+        int longestId = -1; size_t longestLen = 0;
+        for (auto const& [word, id] : wordToId) {
+            if (text.compare(pos, word.length(), word) == 0) {
+                if (word.length() > longestLen) {
+                    longestLen = word.length(); longestId = id;
+                }
+            }
+        }
+        if (longestId != -1) {
+            tokens.push_back(longestId);
+            pos += longestLen;
+        } else pos++;
+    }
+    return tokens;
+}
+
+Embedder::Embedder(int vSize, int dim) {
+    std::mt19937 gen(42);
+    std::uniform_real_distribution<double> dist(-1.0, 1.0);
+    matrix.resize(vSize, std::vector<double>(dim));
+    for(int i=0; i<vSize; i++)
+        for(int j=0; j<dim; j++) matrix[i][j] = dist(gen);
+}
+
+std::vector<double> Embedder::get(int id) {
+    if (id >= 0 && id < (int)matrix.size()) return matrix[id];
+    return std::vector<double>(matrix[0].size(), 0.0);
+}
\ No newline at end of file
diff --git a/Xenith/token/token.h b/Xenith/token/token.h
new file mode 100644
index 0000000..095bb55
--- /dev/null
+++ b/Xenith/token/token.h
@@ -0,0 +1,42 @@
+#ifndef TOKEN_H
+#define TOKEN_H
+
+#include <string>
+#include <vector>
+#include <map>
+
+class Tokenizer {
+public:
+    std::map<std::string, int> wordToId;
+    std::map<int, std::string> idToWord;
+
+    Tokenizer() {
+        add("<EOS>");      // 0
+        add("[SYS]");      // 1
+        add("[USER]");     // 2
+        add("[AI]");       // 3
+        add(" ");          // 4
+        add("\n");         // 5
+        add("привет");     // 6
+        add("как");        // 7
+        add("дела");       // 8
+        add("?");          // 9
+        add("я");          // 10
+        add("робот");      // 11
+        add("хорошо");     // 12
+    }
+
+    void add(std::string word);
+    int getID(std::string word);
+    std::string getWord(int id);
+    std::vector<int> textToTokens(const std::string& text);
+};
+
+class Embedder {
+public:
+    std::vector<std::vector<double>> matrix;
+    Embedder(int vSize, int dim);
+    std::vector<double> get(int id);
+};
+
+#endif
\ No newline at end of file
diff --git a/Xenith/typedef.h b/Xenith/typedef.h
index 31fdc29..348e38c 100644
--- a/Xenith/typedef.h
+++ b/Xenith/typedef.h
@@ -1,15 +1,11 @@
 #ifndef TYPEDEF_H
 #define TYPEDEF_H
 
-#include <vector>
+const int MAX_CONTEXT = 4;   // Сколько токенов видит сеть
+const int EMBED_DIM = 4;     // Размер вектора одного токена
+const int MAX_VOCAB = 13;     // Размер словаря
 
-typedef enum {
-    SIGMOID
-} FunctionActivate_t;
-
-typedef struct {
-    int size;
-    FunctionActivate_t activate;
-} LayerStructure_t;
+typedef enum { SIGMOID } FunctionActivate_t;
+typedef struct { int size; FunctionActivate_t activate; } LayerStructure_t;
 
 #endif
\ No newline at end of file
diff --git a/dataset.txt b/dataset.txt
new file mode 100644
index 0000000..38b9c75
--- /dev/null
+++ b/dataset.txt
@@ -0,0 +1 @@
+[USER]привет[AI]привет как дела?<EOS>
\ No newline at end of file
diff --git a/main b/main
index cbf00c0..a7ad854 100755
Binary files a/main and b/main differ
diff --git a/main.cpp b/main.cpp
index c77f4e6..04e8e20 100644
--- a/main.cpp
+++ b/main.cpp
@@ -1,39 +1,195 @@
 #include <iostream>
 #include <vector>
-#include <ctime>
+#include <string>
+#include <iomanip>
+#include <sstream>
+#include <fstream>
+#include <algorithm>
 #include "Xenith/core.h"
-#include "Xenith/typedef.h"
+#include "Xenith/token/token.h"
 
+// Глобальные настройки
+std::string currentSystemPrompt = "я робот";
 
-int main() {
-    srand(time(NULL));
-
-    LayerStructure_t layers[] = {
-        {2, SIGMOID}, // Вход: 2 числа
-        {3, SIGMOID}, // Скрытый слой
-        {1, SIGMOID}  // Выход: 1 число
-    };
-
-    NeuralNetwork nn(layers, 3);
-
-    // Данные для обучения
-    std::vector<std::vector<double>> inputs = {{1, 1}, {1, 0}, {0, 0}, {0, 1}};
-    std::vector<std::vector<double>> targets = {{0}, {1}, {1}, {0}};
-
-    // Цикл обучения
-    std::cout << "Training..." << std::endl;
-    for (int epoch = 0; epoch < 20000; epoch++) {
-        for (int i = 0; i < inputs.size(); i++) {
-            nn.train(inputs[i], targets[i], 0.5);
-        }
+void printParameterCount(LayerStructure_t layers[], int numLayers) {
+    long long totalParams = 0;
+    for (int i = 0; i < numLayers - 1; i++) {
+        long long weights = (long long)layers[i].size * layers[i + 1].size;
+        long long biases = (long long)layers[i + 1].size;
+        totalParams += (weights + biases);
     }
 
-    // Проверка результатов
-    std::cout << "Results:" << std::endl;
-    for (int i = 0; i < inputs.size(); i++) {
-        std::vector<double> res = nn.feedForward(inputs[i]);
-        std::cout << inputs[i][0] << " " << inputs[i][1] << " -> " 
-                  << (res[0] > 0.5 ? 1 : 0) << " (raw: " << res[0] << ")" << std::endl;
+    std::cout << "--- Xenith AI (Model Size: ";
+    if (totalParams >= 1000000000000LL)
+        std::cout << std::fixed << std::setprecision(1) << (double)totalParams / 1000000000000.0 << "t";
+    else if (totalParams >= 1000000000LL)
+        std::cout << std::fixed << std::setprecision(1) << (double)totalParams / 1000000000.0 << "b";
+    else if (totalParams >= 1000000LL)
+        std::cout << std::fixed << std::setprecision(1) << (double)totalParams / 1000000.0 << "m";
+    else if (totalParams >= 1000LL)
+        std::cout << std::fixed << std::setprecision(1) << (double)totalParams / 1000.0 << "k";
+    else
+        std::cout << totalParams;
+    
+    std::cout << " parameters) ---" << std::endl;
+}
+
+std::vector<double> buildNetInput(const std::vector<int>& tokens, Embedder& emb) {
+    std::vector<double> netInput;
+    netInput.reserve(MAX_CONTEXT * EMBED_DIM);
+    int start = (int)tokens.size() - MAX_CONTEXT;
+    if (start < 0) start = 0;
+    int count = 0;
+    for (int i = start; i < (int)tokens.size(); i++) {
+        std::vector<double> v = emb.get(tokens[i]);
+        netInput.insert(netInput.end(), v.begin(), v.end());
+        count++;
+    }
+    while (count < MAX_CONTEXT) {
+        for (int d = 0; d < EMBED_DIM; d++) netInput.push_back(0.0);
+        count++;
+    }
+    return netInput;
+}
+
+void trainOnSequence(NeuralNetwork& nn, Tokenizer& tok, Embedder& emb, const std::string& dataset, int epochs, double lr) {
+    std::vector<int> allTokens = tok.textToTokens(dataset);
+    if (allTokens.size() < 2) {
+        std::cout << "Error: Sequence too short for training." << std::endl;
+        return;
+    }
+
+    std::cout << "Training logic: Next Token Prediction..." << std::endl;
+    for (int e = 1; e <= epochs; e++) {
+        double totalLoss = 0;
+        for (size_t i = 1; i < allTokens.size(); i++) {
+            std::vector<int> context;
+            for (size_t j = 0; j < i; j++) context.push_back(allTokens[j]);
+            
+            std::vector<double> target(MAX_VOCAB, 0.0);
+            target[allTokens[i]] = 1.0;
+
+            totalLoss += nn.train(buildNetInput(context, emb), target, lr);
+        }
+        std::cout << "\rEpoch " << e << "/" << epochs << " | Loss: " << std::fixed << std::setprecision(6) << totalLoss << std::flush;
+    }
+    std::cout << "\nDone!" << std::endl;
+}
+
+int main() {
+    Tokenizer tok;
+    Embedder emb(MAX_VOCAB, EMBED_DIM);
+    LayerStructure_t layers[] = {
+        {MAX_CONTEXT * EMBED_DIM, SIGMOID}, 
+        {16, SIGMOID}, 
+        {MAX_VOCAB, SIGMOID}
+    };
+    int numLayers = sizeof(layers) / sizeof(layers[0]);
+
+    NeuralNetwork nn(layers, numLayers);
+    printParameterCount(layers, numLayers);
+
+    std::cout << "\n--- MENU ---" << std::endl;
+    std::cout << "/train\n/trainFile\n/help\n/exit\n";
+
+    while (true) {
+        std::cout << "\nxentith~$ ";
+
+        std::string cmdIn;
+        std::getline(std::cin, cmdIn);
+
+        if (cmdIn == "/exit") break;
+
+        if (cmdIn == "/train") {
+            int epochs;
+            double lr;
+            std::cout << "--- Training Setup ---\n";
+            std::cout << "Enter number of epochs: ";
+            std::string epStr; std::getline(std::cin, epStr);
+            epochs = std::stoi(epStr);
+            
+            std::cout << "Enter learning rate (e.g. 0.1): ";
+            std::string lrStr; std::getline(std::cin, lrStr);
+            lr = std::stod(lrStr);
+
+            std::cout << "\n--- Example Content ---\n";
+            std::cout << "User: ";
+            std::string userPart;
+            std::getline(std::cin, userPart);
+
+            std::cout << "AI: ";
+            std::string aiPart;
+            std::getline(std::cin, aiPart);
+
+            std::string finalData = "[SYS]" + currentSystemPrompt + 
+                                    "[USER]" + userPart + 
+                                    "[AI]" + aiPart + "<EOS>";
+
+            std::cout << "\nTraining logic: Pattern Recognition..." << std::endl;
+            trainOnSequence(nn, tok, emb, finalData, epochs, lr);
+        }
+
+        else if (cmdIn == "/trainFile") {
+            std::string content;
+            std::cout << "Enter filename: ";
+            std::string filename;
+            std::getline(std::cin, filename);
+            std::ifstream file(filename);
+            if (file.is_open()) {
+                std::stringstream buffer;
+                buffer << file.rdbuf();
+                content = buffer.str();
+                std::cout << "Loaded " << content.length() << " characters from file." << std::endl;
+            } else {
+                std::cout << "Could not open file!" << std::endl;
+                continue;
+            }
+
+            int epochs;
+            double lr;
+            std::cout << "Enter number of epochs: ";
+            std::string epStr; std::getline(std::cin, epStr);
+            epochs = std::stoi(epStr);
+            
+            std::cout << "Enter learning rate (e.g. 0.1): ";
+            std::string lrStr; std::getline(std::cin, lrStr);
+            lr = std::stod(lrStr);
+
+            std::string finalData = "[SYS]" + currentSystemPrompt + content + "<EOS>";
+            trainOnSequence(nn, tok, emb, finalData, epochs, lr);
+
+        } else if (cmdIn == "/sysPrompt") {
+            std::cout << "Current System Prompt: " << currentSystemPrompt << std::endl;
+            std::cout << "Enter new System Prompt: ";
+            std::getline(std::cin, currentSystemPrompt);
+            std::cout << "System Prompt updated!" << std::endl;
+
+        } else if (cmdIn == "/help") {
+            std::cout << "\n--- MENU ---" << std::endl;
+            std::cout << "/train\n/trainFile\n/sysPrompt\n/help\n/exit\n";
+
+        } else {
+            std::string prompt = "[SYS]" + currentSystemPrompt + "[USER]" + cmdIn + "[AI]";
+            std::vector<int> currentTokens = tok.textToTokens(prompt);
+
+            std::cout << "AI: ";
+            for (int g = 0; g < 30; g++) {
+                std::vector<double> out = nn.feedForward(buildNetInput(currentTokens, emb));
+                
+                int bestId = 0;
+                for (int i = 0; i < MAX_VOCAB; i++) {
+                    if (out[i] > out[bestId]) bestId = i;
+                }
+
+                if (bestId == 0) break;
+                
+                std::string word = tok.getWord(bestId);
+                std::cout << word << std::flush;
+                
+                currentTokens.push_back(bestId);
+            }
+            std::cout << std::endl;
+        }
     }
 
     return 0;