gpt-driver-node 1.0.0-alpha.8 → 1.0.0

package/dist/index.cjs

@@ -1,8 +1,13 @@
 'use strict';
 
+var node_fs = require('node:fs');
+var path = require('node:path');
 var axios = require('axios');
-var seleniumWebdriver = require('selenium-webdriver');
 var sharp = require('sharp');
+var webdriverio = require('webdriverio');
+var winston = require('winston');
+var zod = require('zod');
+var crypto = require('node:crypto');
 
 const delay = async (milliseconds) => {
   await new Promise((resolve) => setTimeout(resolve, milliseconds));
@@ -18,12 +23,600 @@ function buildUrl(base, extraPath) {
   return `${baseUrl}${extraPath}`;
 }
 
+const colors = {
+  reset: "\x1B[0m",
+  bold: "\x1B[1m",
+  gray: "\x1B[90m",
+  red: "\x1B[31m",
+  green: "\x1B[32m",
+  yellow: "\x1B[33m",
+  cyan: "\x1B[36m"
+};
+const logStyles = {
+  bold: (text) => `${colors.bold}${text}${colors.reset}`,
+  cyan: (text) => `${colors.cyan}${text}${colors.reset}`,
+  yellow: (text) => `${colors.yellow}${text}${colors.reset}`,
+  green: (text) => `${colors.green}${text}${colors.reset}`,
+  red: (text) => `${colors.red}${text}${colors.reset}`,
+  gray: (text) => `${colors.gray}${text}${colors.reset}`,
+  highlight: (text) => `${colors.bold}${colors.cyan}${text}${colors.reset}`,
+  success: (text) => `${colors.bold}${colors.green}${text}${colors.reset}`,
+  error: (text) => `${colors.bold}${colors.red}${text}${colors.reset}`,
+  warning: (text) => `${colors.bold}${colors.yellow}${text}${colors.reset}`
+};
+const globalLogger = winston.createLogger({
+  level: process.env.GPT_DRIVER_LOG_LEVEL || "info",
+  format: winston.format.combine(
+    winston.format.timestamp({ format: "YYYY-MM-DD HH:mm:ss" }),
+    winston.format.errors({ stack: true }),
+    winston.format.printf(({ timestamp, level, message, stack }) => {
+      const logMessage = `${timestamp} [${level.toUpperCase()}]: ${message}`;
+      return stack ? `${logMessage}
+${stack}` : logMessage;
+    })
+  ),
+  transports: [
+    new winston.transports.Console({
+      format: winston.format.combine(
+        winston.format.printf(({ timestamp, level, message, stack }) => {
+          let coloredLevel = level.toUpperCase();
+          switch (level) {
+            case "error":
+              coloredLevel = logStyles.error(coloredLevel);
+              break;
+            case "warn":
+              coloredLevel = logStyles.warning(coloredLevel);
+              break;
+            case "info":
+              coloredLevel = logStyles.cyan(coloredLevel);
+              break;
+            case "debug":
+              coloredLevel = logStyles.gray(coloredLevel);
+              break;
+          }
+          const formattedTimestamp = logStyles.gray(timestamp);
+          const logMessage = `${formattedTimestamp} [${coloredLevel}]: ${message}`;
+          return stack ? `${logMessage}
+${logStyles.gray(stack)}` : logMessage;
+        })
+      )
+    })
+  ]
+});
+
+const SavableStepBaseSchema = zod.z.object({
+  id: zod.z.number().optional(),
+  descriptionText: zod.z.string().optional(),
+  optional: zod.z.boolean().optional()
+});
+const SavableTapStepSchema = SavableStepBaseSchema.extend({
+  type: zod.z.literal("tap"),
+  elementId: zod.z.string().optional(),
+  timeout: zod.z.number().optional(),
+  useLlmOnly: zod.z.boolean().optional(),
+  cropBase64: zod.z.string().optional()
+});
+const SavableAssertStepSchema = SavableStepBaseSchema.extend({
+  type: zod.z.literal("assert"),
+  elementId: zod.z.string().optional(),
+  timeout: zod.z.number().optional(),
+  useLlmOnly: zod.z.boolean().optional(),
+  cropBase64: zod.z.string().optional()
+});
+const SavableTypeStepSchema = SavableStepBaseSchema.extend({
+  type: zod.z.literal("type"),
+  text: zod.z.string()
+});
+const SavableScrollStepSchema = SavableStepBaseSchema.extend({
+  type: zod.z.literal("scroll"),
+  direction: zod.z.enum(["up", "down"])
+});
+const SavableZoomStepSchema = SavableStepBaseSchema.extend({
+  type: zod.z.literal("zoom"),
+  direction: zod.z.enum(["in", "out"])
+});
+const SavableScrollUntilStepSchema = SavableStepBaseSchema.extend({
+  type: zod.z.literal("scrollUntil"),
+  text: zod.z.string().optional(),
+  elementId: zod.z.string().optional(),
+  direction: zod.z.enum(["up", "down"]),
+  maxScrolls: zod.z.number().optional()
+});
+const SavableDeeplinkStepSchema = SavableStepBaseSchema.extend({
+  type: zod.z.literal("deeplink"),
+  url: zod.z.string()
+});
+const SavableAIStepSchema = SavableStepBaseSchema.extend({
+  type: zod.z.literal("ai"),
+  instruction: zod.z.string()
+});
+const SavableFileRefStepSchema = SavableStepBaseSchema.extend({
+  type: zod.z.literal("fileRef"),
+  path: zod.z.string(),
+  overrides: zod.z.record(zod.z.string(), zod.z.string()).optional()
+});
+const SavableStepSchema = zod.z.discriminatedUnion("type", [
+  SavableTapStepSchema,
+  // type: 'tap'
+  SavableAssertStepSchema,
+  // type: 'assert'
+  SavableTypeStepSchema,
+  // type: 'type'
+  SavableScrollStepSchema,
+  // type: 'scroll'
+  SavableZoomStepSchema,
+  // type: 'zoom'
+  SavableScrollUntilStepSchema,
+  // type: 'scrollUntil'
+  SavableDeeplinkStepSchema,
+  // type: 'deeplink'
+  SavableAIStepSchema,
+  // type: 'ai'
+  SavableFileRefStepSchema
+  // type: 'fileRef'
+]);
+const SavableTestStoreSchema = zod.z.object({
+  name: zod.z.string(),
+  steps: zod.z.array(SavableStepSchema),
+  params: zod.z.record(zod.z.string(), zod.z.string()).optional()
+});
+
+const CACHE_SERVER_URL = "https://cache.mobileboost.io";
+const GPT_DRIVER_BASE_URL = "https://api.mobileboost.io";
+const RESCALE_FACTOR = 4;
+const SMART_LOOP_MAX_ITERATIONS = 15;
+const CACHE_RETRY_MS = 2e3;
+const CACHE_CHECK_INTERVAL_MS = 500;
+
+function generateCacheHash(apiKey, filepath, stepNumber, description, platform, resolution) {
+  const resString = resolution ? `${resolution.width}x${resolution.height}` : "";
+  const normalizedPlatform = platform?.toLowerCase() || "";
+  const data = `${apiKey}${filepath || ""}${stepNumber}${description}${normalizedPlatform || ""}${resString}`;
+  return crypto.createHash("sha256").update(data).digest("hex");
+}
+function scaleCommand(cmd, operation) {
+  if (cmd.match(/([xy])=(\d+)/)) {
+    return cmd.replace(/([xy])=(\d+)/g, (_match, axis, val) => {
+      const num = parseInt(val, 10);
+      let scaled;
+      if (operation === "multiply") {
+        scaled = Math.round(num * RESCALE_FACTOR);
+      } else {
+        scaled = Math.round(num / RESCALE_FACTOR);
+      }
+      return `${axis}=${scaled}`;
+    });
+  }
+  return cmd.replace(/(^|;)(\d+);(\d+)(;|$)/, (_match, prefix, xStr, yStr, suffix) => {
+    const x = parseInt(xStr, 10);
+    const y = parseInt(yStr, 10);
+    let scaledX;
+    let scaledY;
+    if (operation === "multiply") {
+      scaledX = Math.round(x * RESCALE_FACTOR);
+      scaledY = Math.round(y * RESCALE_FACTOR);
+    } else {
+      scaledX = Math.round(x / RESCALE_FACTOR);
+      scaledY = Math.round(y / RESCALE_FACTOR);
+    }
+    return `${prefix}${scaledX};${scaledY}${suffix}`;
+  });
+}
+async function resizeScreenshotForCache(screenshotBase64) {
+  const buffer = Buffer.from(
+    screenshotBase64.replace(/^data:image\/\w+;base64,/, ""),
+    "base64"
+  );
+  const metadata = await sharp(buffer).metadata();
+  const originalWidth = metadata.width ?? 1080;
+  const desiredWidth = Math.round(originalWidth / RESCALE_FACTOR);
+  return sharp(buffer).resize({ width: desiredWidth, withoutEnlargement: true }).toBuffer();
+}
+
+async function executeFromCache(params) {
+  try {
+    const hash = generateCacheHash(
+      params.apiKey,
+      params.filepath,
+      params.stepNumber,
+      params.stepDescription,
+      params.platform,
+      params.screenResolution
+    );
+    const resizedBuffer = await resizeScreenshotForCache(params.screenshot);
+    const formData = new FormData();
+    formData.append("hash", hash);
+    const blob = new Blob([new Uint8Array(resizedBuffer)], { type: "image/png" });
+    const blobSizeMB = (blob.size / (1024 * 1024)).toFixed(2);
+    globalLogger.debug(`[Cache] Executing from cache with screenshot size: ${blobSizeMB} MB`);
+    formData.append("screenshot", blob, "screenshot.png");
+    if (params.highestUsedIndex !== void 0 && params.highestUsedIndex !== null) {
+      globalLogger.debug(`[Cache] Sending highest_used_index: ${params.highestUsedIndex}`);
+      formData.append("highest_used_index", String(params.highestUsedIndex));
+    }
+    const response = await axios.post(`${CACHE_SERVER_URL}/execute-from-cache`, formData);
+    const result = response.data;
+    if (result.found && result.cacheCommands) {
+      const scaledCommands = result.cacheCommands.map(
+        (cmd) => scaleCommand(cmd, "multiply")
+      );
+      return {
+        found: true,
+        cacheCommands: scaledCommands,
+        cacheIndex: result.cacheIndex
+      };
+    }
+    return { found: false };
+  } catch (error) {
+    if (axios.isAxiosError(error)) {
+      globalLogger.warn(`[Cache] Cache lookup failed: ${error.response?.data || error.message}`);
+    } else {
+      globalLogger.error(`[Cache] Error executing from cache: ${error}`);
+    }
+    return { found: false };
+  }
+}
+async function populateCache(params) {
+  try {
+    const hash = generateCacheHash(
+      params.apiKey,
+      params.filepath,
+      params.stepNumber,
+      params.stepDescription,
+      params.platform,
+      params.screenResolution
+    );
+    const payload = await Promise.all(params.executionData.map(async (item) => {
+      const resizedBuffer = await resizeScreenshotForCache(item.screenshot);
+      const scaledCommands = item.commands.map(
+        (cmd) => scaleCommand(cmd, "divide")
+      );
+      return {
+        screenshot: resizedBuffer.toString("base64"),
+        commands: scaledCommands
+      };
+    }));
+    const payloadSizeMB = (JSON.stringify(payload).length / (1024 * 1024)).toFixed(2);
+    globalLogger.debug(`[Cache] Populating cache with payload size: ~${payloadSizeMB} MB (Hash: ${hash})`);
+    await axios.post(`${CACHE_SERVER_URL}/populate-cache`, payload, {
+      params: { hash }
+    });
+    return { success: true };
+  } catch (error) {
+    if (axios.isAxiosError(error)) {
+      globalLogger.error(`[Cache] Failed to populate cache: ${error.response?.data || error.message}`);
+    } else {
+      globalLogger.error(`[Cache] Error populating cache: ${error}`);
+    }
+    return { success: false };
+  }
+}
+
+const AI_AGENT_ENDPOINT = "https://api.mobileboost.io/call_lambda";
+async function executeAgentStep(params) {
+  const imageBuffer = Buffer.from(params.base64_screenshot, "base64");
+  const metadata = await sharp(imageBuffer).metadata();
+  const originalWidth = metadata.width ?? 1080;
+  const originalHeight = metadata.height ?? 1920;
+  const desiredWidth = Math.round(originalWidth / RESCALE_FACTOR);
+  const resizedBuffer = await sharp(imageBuffer).resize({ width: desiredWidth, withoutEnlargement: true }).toBuffer();
+  const resizedMetadata = await sharp(resizedBuffer).metadata();
+  const resizedWidth = resizedMetadata.width ?? desiredWidth;
+  const resizedHeight = resizedMetadata.height ?? Math.round(originalHeight * (desiredWidth / originalWidth));
+  globalLogger.debug(`[AI Client] Resized screenshot: ${originalWidth}x${originalHeight} -> ${resizedWidth}x${resizedHeight}`);
+  const payload = {
+    lambda_flow: "get_next_step",
+    current_date: (/* @__PURE__ */ new Date()).toLocaleDateString("en-GB", {
+      day: "numeric",
+      month: "long",
+      year: "numeric"
+    }),
+    base64_screenshot: resizedBuffer.toString("base64"),
+    getUI_elements: [],
+    uiHierarchy: [],
+    test_task_string: JSON.stringify([
+      {
+        id: "step-1",
+        text: `1. ${params.instruction}`,
+        plainText: params.instruction
+      }
+    ]),
+    image_width: resizedWidth,
+    image_height: resizedHeight,
+    action_history: params.action_history,
+    orgKey: params.apiKey,
+    template_images: {},
+    model_provider: "vellum",
+    model_version: "claude-agent",
+    fallbackModel: "claude-agent",
+    utilize_fullTextAnnotation: false,
+    enableSortingOCR: true,
+    enableActionHistoryCut: true,
+    removeOverlappingText: false,
+    currentAndPreviousScreenMatch: false,
+    popupDetectionEnabled: true,
+    ocrProvider: "gcp"
+  };
+  globalLogger.debug(`[AI Client] Sending request to ${AI_AGENT_ENDPOINT}`);
+  try {
+    const response = await axios.post(
+      AI_AGENT_ENDPOINT,
+      payload,
+      {
+        headers: {
+          "Content-Type": "application/json"
+        }
+      }
+    );
+    const result = response.data;
+    globalLogger.debug("[AI Client] Received response from backend");
+    if (result.appetizeCommands) {
+      result.appetizeCommands = result.appetizeCommands.map(
+        (cmd) => scaleCommand(cmd, "multiply")
+      );
+    }
+    return result;
+  } catch (error) {
+    if (axios.isAxiosError(error)) {
+      const status = error.response?.status ?? "unknown";
+      const errorText = error.response?.data ?? error.message;
+      globalLogger.error(`[AI Client] Backend error (${status}): ${JSON.stringify(errorText)}`);
+      throw new Error(`AI Backend Error: ${status} - ${error.message}`);
+    }
+    throw error;
+  }
+}
+
+function parseTapCoordinates(cmd) {
+  const xMatch = cmd.match(/x=(\d+)/);
+  const yMatch = cmd.match(/y=(\d+)/);
+  if (xMatch && yMatch) {
+    return {
+      x: parseInt(xMatch[1], 10),
+      y: parseInt(yMatch[1], 10)
+    };
+  }
+  const parts = cmd.split(";");
+  if (parts.length >= 3) {
+    const x = parseInt(parts[1], 10);
+    const y = parseInt(parts[2], 10);
+    if (!isNaN(x) && !isNaN(y)) {
+      return { x, y };
+    }
+  }
+  return null;
+}
+function parseWaitSeconds(cmd) {
+  const match = cmd.match(/wait:\s*(\d+)/);
+  return match ? parseInt(match[1], 10) : null;
+}
+function parseScrollDirection(cmd) {
+  const match = cmd.match(/scroll:\s*(up|down)/i);
+  return match ? match[1].toLowerCase() : null;
+}
+function parseTypeText(cmd) {
+  const match = cmd.match(/^type:\s*(.+)$/);
+  return match ? match[1] : null;
+}
+function isTaskComplete(cmd) {
+  return cmd.toLowerCase().includes("task complete:");
+}
+function isErrorDetected(cmd) {
+  return cmd.toLowerCase().includes("error detected:");
+}
+function isRememberCommand(cmd) {
+  return cmd.startsWith("remember:");
+}
+function isTapCommand(cmd) {
+  return /^t(ap|ab)On:/.test(cmd);
+}
+function isWaitCommand(cmd) {
+  return cmd.startsWith("wait:");
+}
+function isScrollCommand(cmd) {
+  return cmd.startsWith("scroll:");
+}
+function isTypeCommand(cmd) {
+  return cmd.startsWith("type:");
+}
+
+async function executeSmartLoop(ctx, params) {
+  const maxCacheAttempts = Math.floor(CACHE_RETRY_MS / CACHE_CHECK_INTERVAL_MS);
+  let loopCount = 0;
+  let actionHistory = [...ctx.globalActionHistory];
+  let lastCacheIndex = void 0;
+  let anyCacheMiss = false;
+  let everHadCacheHit = false;
+  const currentExecutionData = [];
+  globalLogger.info(`[SmartLoop] Starting for step ${params.stepNumber}: "${params.description}"`);
+  try {
+    while (loopCount < SMART_LOOP_MAX_ITERATIONS) {
+      let screenshot = "";
+      let commands = [];
+      let isCacheHit = false;
+      for (let attempt = 0; attempt < maxCacheAttempts; attempt++) {
+        screenshot = await ctx.getScreenshot();
+        const sizeInBytes = screenshot.length * 0.75;
+        const sizeInMB = (sizeInBytes / (1024 * 1024)).toFixed(2);
+        globalLogger.debug(`[SmartLoop] Captured screenshot: ~${sizeInMB} MB`);
+        try {
+          globalLogger.debug(`[SmartLoop] Checking cache (Attempt ${attempt + 1}/${maxCacheAttempts})`);
+          const cacheResult = await executeFromCache({
+            apiKey: ctx.apiKey,
+            stepNumber: params.stepNumber,
+            stepDescription: params.description,
+            screenshot,
+            screenResolution: ctx.screenSize,
+            highestUsedIndex: lastCacheIndex,
+            platform: ctx.platform,
+            filepath: params.filepath
+          });
+          if (cacheResult.found && cacheResult.cacheCommands) {
+            commands = cacheResult.cacheCommands;
+            lastCacheIndex = cacheResult.cacheIndex;
+            isCacheHit = true;
+            everHadCacheHit = true;
+            globalLogger.info(`[SmartLoop] Cache Hit! (${commands.length} commands)`);
+            break;
+          }
+        } catch (e) {
+          globalLogger.warn(`[SmartLoop] Cache check failed: ${e.message}`);
+        }
+        if (attempt < maxCacheAttempts - 1) {
+          globalLogger.debug(`[SmartLoop] Cache miss, retrying in ${CACHE_CHECK_INTERVAL_MS}ms...`);
+          await delay(CACHE_CHECK_INTERVAL_MS);
+        }
+      }
+      let aiCommands = [];
+      if (!isCacheHit) {
+        anyCacheMiss = true;
+        globalLogger.info(`[SmartLoop] Cache Miss. Requesting AI agent...`);
+        const agentResponse = await executeAgentStep({
+          apiKey: ctx.apiKey,
+          base64_screenshot: screenshot,
+          instruction: params.instruction,
+          action_history: actionHistory
+        });
+        aiCommands = agentResponse.appetizeCommands || [];
+        const gptCommands = agentResponse.gptCommands || [];
+        const reasoningIndex = gptCommands.findIndex((entry) => entry.startsWith("reasoning:"));
+        if (reasoningIndex !== -1) {
+          const parsedCommands = gptCommands.slice(reasoningIndex);
+          const rememberCommands = parsedCommands.filter((cmd) => isRememberCommand(cmd));
+          if (rememberCommands.length > 0) {
+            ctx.globalActionHistory.push(...rememberCommands);
+          }
+          actionHistory = [...actionHistory, ...parsedCommands];
+        }
+        commands = [...aiCommands];
+        globalLogger.debug(`[SmartLoop] AI returned ${commands.length} command(s)`);
+      }
+      currentExecutionData.push({
+        screenshot,
+        commands: aiCommands.length > 0 ? aiCommands : commands
+      });
+      await ctx.logCodeExecution(screenshot, commands.join("\n"));
+      let actionExecuted = false;
+      let taskCompleted = false;
+      if (commands.length > 0) {
+        globalLogger.debug(`[SmartLoop] Executing ${commands.length} command(s)`);
+      }
+      for (const cmd of commands) {
+        if (isTaskComplete(cmd)) {
+          taskCompleted = true;
+          globalLogger.info(`[SmartLoop] Task completed signal received`);
+          continue;
+        }
+        if (isErrorDetected(cmd)) {
+          throw new Error(`AI Reported Error: ${cmd}`);
+        }
+        if (isRememberCommand(cmd)) {
+          ctx.globalActionHistory.push(cmd);
+        }
+        if (isTapCommand(cmd)) {
+          const coords = parseTapCoordinates(cmd);
+          if (coords) {
+            globalLogger.debug(`[SmartLoop] Executing tap at (${coords.x}, ${coords.y})`);
+            await ctx.performTap(coords.x, coords.y);
+            actionExecuted = true;
+          }
+        } else if (isWaitCommand(cmd)) {
+          const seconds = parseWaitSeconds(cmd);
+          if (seconds) {
+            globalLogger.debug(`[SmartLoop] Waiting ${seconds}s`);
+            await delay(seconds * 1e3);
+            actionExecuted = true;
+          }
+        } else if (isScrollCommand(cmd)) {
+          const direction = parseScrollDirection(cmd);
+          if (direction) {
+            globalLogger.debug(`[SmartLoop] Scrolling ${direction}`);
+            await ctx.performScroll(direction);
+            actionExecuted = true;
+          }
+        } else if (isTypeCommand(cmd)) {
+          const text = parseTypeText(cmd);
+          if (text) {
+            globalLogger.debug(`[SmartLoop] Typing text`);
+            await ctx.performType(text);
+            actionExecuted = true;
+          }
+        }
+      }
+      if (actionExecuted) {
+        if (isCacheHit) {
+          actionHistory.push(...commands);
+        }
+        await delay(100);
+      }
+      if (taskCompleted) {
+        globalLogger.info(`[SmartLoop] Task completed successfully`);
+        if (anyCacheMiss && currentExecutionData.length > 0) {
+          globalLogger.info(`[SmartLoop] Populating cache with ${currentExecutionData.length} frame(s)...`);
+          try {
+            await populateCache({
+              apiKey: ctx.apiKey,
+              stepNumber: params.stepNumber,
+              stepDescription: params.description,
+              executionData: currentExecutionData,
+              screenResolution: ctx.screenSize,
+              platform: ctx.platform,
+              filepath: params.filepath
+            });
+            globalLogger.debug(`[SmartLoop] Cache populated successfully`);
+          } catch (e) {
+            globalLogger.warn(`[SmartLoop] Failed to populate cache: ${e.message}`);
+          }
+        } else if (!anyCacheMiss) {
+          globalLogger.debug(`[SmartLoop] Skipping cache population (all actions were cached)`);
+        }
+        return {
+          success: true,
+          iterations: loopCount + 1,
+          cacheHit: everHadCacheHit
+        };
+      }
+      loopCount++;
+    }
+    throw new Error(`Smart Loop timeout after ${SMART_LOOP_MAX_ITERATIONS} iterations`);
+  } catch (error) {
+    const message = error instanceof Error ? error.message : String(error);
+    globalLogger.error(`[SmartLoop] Error: ${message}`);
+    return {
+      success: false,
+      error: message,
+      iterations: loopCount + 1,
+      cacheHit: everHadCacheHit
+    };
+  }
+}
+
 class GptDriver {
+  interpolateTemplate(input, params) {
+    if (typeof input !== "string" || !input.includes("{{")) return input;
+    const pattern = /{{\s*([^}]+?)\s*}}/g;
+    return input.replace(pattern, (_match, keyRaw) => {
+      const key = String(keyRaw);
+      if (!(key in params)) {
+        throw new Error(`Missing flow param: {{${key}}}`);
+      }
+      return params[key];
+    });
+  }
   apiKey;
   gptDriverSessionId;
   gptDriverBaseUrl;
   appiumSessionConfig;
+  cachingMode;
   driver;
+  appiumSessionStarted;
+  useGptDriverCloud;
+  gptDriverCloudConfig;
+  buildId;
+  testId;
+  step_number = 1;
+  // Smart loop state - maintains action history across steps for context
+  globalActionHistory = [];
   /**
    * Creates an instance of the GptDriver class.
    *
@@ -43,19 +636,34 @@ class GptDriver {
    *                 - `device.platform`: The platform name of the device (e.g., iOS, Android).
    */
   constructor(config) {
+    this.testId = config.testId;
     this.apiKey = config.apiKey;
-    this.gptDriverBaseUrl = "https://api.mobileboost.io";
-    this.initializeDriver(config);
-    this.initializeAppiumConfig(config);
+    this.buildId = config.buildId;
+    this.useGptDriverCloud = config.useGptDriverCloud;
+    this.gptDriverBaseUrl = GPT_DRIVER_BASE_URL;
+    this.cachingMode = config.cachingMode ?? "NONE";
+    if (config.useGptDriverCloud) {
+      if (config.serverConfig.device?.platform == null) {
+        throw new Error("Platform is missing. Please specify the platform when using GPTDriver Cloud.");
+      }
+      this.gptDriverCloudConfig = {
+        platform: config.serverConfig.device.platform,
+        deviceName: config.serverConfig.device.deviceName,
+        platformVersion: config.serverConfig.device.platformVersion
+      };
+    } else {
+      this.initializeDriver(config);
+      this.initializeAppiumConfig(config);
+    }
   }
   initializeDriver(config) {
     if (config.driver) {
       this.driver = config.driver;
-      if (!config.severConfig?.url) {
+      if (!config.serverConfig.url) {
         throw new Error("Server url is missing. Please specify the server url when providing a driver.");
       }
     } else {
-      const isValidServerConfig = config.severConfig?.url && config.severConfig.device?.platform;
+      const isValidServerConfig = config.serverConfig.url && config.serverConfig.device?.platform;
       if (!isValidServerConfig) {
         throw new Error("Either provide a driver, or a valid severConfig object.");
       }
@@ -64,10 +672,10 @@ class GptDriver {
   initializeAppiumConfig(config) {
     const defaultPort = parseInt(process.env.APPIUM_PORT ?? "4723", 10);
     const defaultHost = process.env.APPIUM_HOST ?? "127.0.0.1";
-    let serverUrl = config.severConfig?.url instanceof URL ? config.severConfig.url : new URL(config.severConfig?.url ?? `http://${defaultHost}:${defaultPort}`);
+    const serverUrl = config.serverConfig.url instanceof URL ? config.serverConfig.url : new URL(config.serverConfig.url ?? `http://${defaultHost}:${defaultPort}`);
     this.appiumSessionConfig = {
       serverUrl,
-      ...config.severConfig?.device
+      ...config.serverConfig.device
     };
   }
   /**
@@ -77,48 +685,53 @@ class GptDriver {
    * @throws {Error} If the session cannot be started or the driver is not properly initialized.
    */
   async startSession() {
-    console.log(">> Starting session...");
-    if (this.driver) {
-      let platform;
-      let platformVersion;
-      let deviceName;
-      let sessionId;
-      if (this.driver instanceof seleniumWebdriver.WebDriver) {
-        const capabilities = await this.driver.getCapabilities();
-        platform = capabilities.get("platformName");
-        platformVersion = capabilities.get("platformVersion") ?? this.appiumSessionConfig?.platformVersion;
-        deviceName = this.appiumSessionConfig?.deviceName ?? capabilities.get("deviceName");
-        const session = await this.driver.getSession();
-        sessionId = session.getId();
+    globalLogger.info("Starting session...");
+    if (!this.useGptDriverCloud) {
+      if (this.driver) {
+        let platform;
+        let platformVersion;
+        let deviceName;
+        let sessionId;
+        if (this.driver.sessionId == null) {
+          const driver = this.driver;
+          const capabilities = await driver.getCapabilities();
+          platform = capabilities.get("platformName");
+          platformVersion = capabilities.get("platformVersion") ?? this.appiumSessionConfig?.platformVersion;
+          deviceName = this.appiumSessionConfig?.deviceName ?? capabilities.get("deviceName");
+          const session = await driver.getSession();
+          sessionId = session.getId();
+        } else {
+          const driver = this.driver;
+          platform = driver.capabilities["appium:platformName"] ?? driver.capabilities["platformName"];
+          platformVersion = driver.capabilities["appium:platformVersion"] ?? driver.capabilities["platformVersion"];
+          deviceName = this.appiumSessionConfig?.deviceName ?? driver.capabilities["appium:deviceName"] ?? driver.capabilities["deviceName"];
+          sessionId = driver.sessionId;
+        }
+        this.appiumSessionConfig = {
+          ...this.appiumSessionConfig,
+          id: sessionId,
+          platform,
+          platformVersion,
+          deviceName
+        };
+        globalLogger.debug(`Session config: ${JSON.stringify(this.appiumSessionConfig)}`);
       } else {
-        platform = this.driver.capabilities["appium:platformName"];
-        platformVersion = this.driver.capabilities["appium:platformVersion"];
-        deviceName = this.appiumSessionConfig?.deviceName ?? this.driver.capabilities["appium:deviceName"] ?? "";
-        sessionId = this.driver.sessionId;
-      }
-      this.appiumSessionConfig = {
-        ...this.appiumSessionConfig,
-        id: sessionId,
-        platform,
-        platformVersion,
-        deviceName
+        this.appiumSessionConfig.id = await this.createSession();
+      }
+      const url = buildUrl(this.appiumSessionConfig.serverUrl, `/session/${this.appiumSessionConfig.id}/window/rect`);
+      const rectResponse = await axios.get(url);
+      this.appiumSessionConfig.size = {
+        width: rectResponse.data.value.width,
+        height: rectResponse.data.value.height
       };
-    } else {
-      this.appiumSessionConfig.id = await this.createSession();
+      this.appiumSessionStarted = true;
     }
     await this.createGptDriverSession();
-    const url = buildUrl(this.appiumSessionConfig.serverUrl, `/session/${this.appiumSessionConfig.id}/window/rect`);
-    const rectResponse = await axios.get(
-      url
-    );
-    this.appiumSessionConfig.size = {
-      width: rectResponse.data.value.width,
-      height: rectResponse.data.value.height
-    };
-    console.log(`>> Session created. Monitor execution at: ${this.getSessionLink()}`);
+    globalLogger.info(logStyles.highlight(`Session created. Monitor execution at: ${this.getSessionLink()}`));
   }
   async createSession() {
     const { platform, deviceName, platformVersion, serverUrl } = this.appiumSessionConfig;
+    globalLogger.debug(`Creating Appium session for ${platform} ${platformVersion} on ${deviceName}`);
     const url = buildUrl(serverUrl, `/session`);
     const response = await axios.post(
       url,
@@ -133,22 +746,41 @@ class GptDriver {
         }
       }
     );
-    return response.data.value.sessionId;
+    const sessionId = response.data.value.sessionId;
+    globalLogger.debug(`Appium session created with ID: ${sessionId}`);
+    return sessionId;
   }
   async createGptDriverSession() {
+    globalLogger.debug("Creating GPT Driver session...");
     const response = await axios.post(
       `${this.gptDriverBaseUrl}/sessions/create`,
       {
+        ...this.testId && { test_id: this.testId },
         api_key: this.apiKey,
-        appium_session_id: this.appiumSessionConfig.id,
+        appium_session_id: this.appiumSessionConfig?.id,
         device_config: {
-          platform: this.appiumSessionConfig.platform,
-          device: this.appiumSessionConfig.deviceName,
-          os: this.appiumSessionConfig.platformVersion
-        }
+          platform: this.appiumSessionConfig?.platform ?? this.gptDriverCloudConfig?.platform,
+          device: this.appiumSessionConfig?.deviceName ?? this.gptDriverCloudConfig?.deviceName,
+          os: this.appiumSessionConfig?.platformVersion ?? this.gptDriverCloudConfig?.platformVersion
+        },
+        use_internal_virtual_device: this.useGptDriverCloud,
+        build_id: this.buildId,
+        caching_mode: this.cachingMode
       }
     );
     this.gptDriverSessionId = response.data.sessionId;
+    globalLogger.debug(`GPT Driver session created with ID: ${this.gptDriverSessionId}`);
+    if (this.useGptDriverCloud) {
+      const parsedUrl = new URL(response.data.appiumServerUrl);
+      this.driver = await webdriverio.attach({
+        options: {
+          hostname: parsedUrl.hostname,
+          path: parsedUrl.pathname
+        },
+        sessionId: response.data.appiumSessionId
+      });
+      this.appiumSessionStarted = true;
+    }
   }
   getSessionLink() {
     return `https://app.mobileboost.io/gpt-driver/sessions/${this.gptDriverSessionId}`;
@@ -164,20 +796,174 @@ class GptDriver {
    *
    * @throws {Error} If the request to stop the session fails.
    */
-  async stopSession(status) {
-    console.log(">> Stopping session...");
-    await axios.post(
-      `${this.gptDriverBaseUrl}/sessions/${this.gptDriverSessionId}/stop`,
+  async setSessionStatus(status) {
+    if (this.gptDriverSessionId) {
+      globalLogger.info(`Stopping session with status: ${status}`);
+      await axios.post(
+        `${this.gptDriverBaseUrl}/sessions/${this.gptDriverSessionId}/stop`,
+        {
+          api_key: this.apiKey,
+          status
+        }
+      );
+      globalLogger.info("Session stopped successfully");
+      this.appiumSessionStarted = false;
+      this.gptDriverSessionId = void 0;
+      this.step_number = 1;
+      this.globalActionHistory = [];
+    }
+  }
+  // ─────────────────────────────────────────────────────────────────────────────
+  // SMART LOOP INTEGRATION
+  // ─────────────────────────────────────────────────────────────────────────────
+  /**
+   * Creates a SmartLoopContext for the current session.
+   * This context provides all the callbacks needed by the smart loop executor.
+   */
+  createSmartLoopContext() {
+    return {
+      apiKey: this.apiKey,
+      platform: this.appiumSessionConfig?.platform,
+      screenSize: this.appiumSessionConfig.size,
+      globalActionHistory: this.globalActionHistory,
+      getScreenshot: () => this.getScreenshot(this.appiumSessionConfig),
+      performTap: (x, y) => this.performTap(x, y),
+      performScroll: (direction) => this.performScroll(direction),
+      performType: (text) => this.performType(text),
+      logCodeExecution: async (screenshot, command) => this.logCodeExecution(screenshot, command)
+    };
+  }
+  /**
+   * Calls the AI agent to determine the next actions based on the current screenshot.
+   * This requires an active GPT Driver session.
+   */
+  async executeAgentStep(params) {
+    const response = await axios.post(
+      `${this.gptDriverBaseUrl}/sessions/${this.gptDriverSessionId}/agent/execute`,
       {
         api_key: this.apiKey,
-        status
+        base64_screenshot: params.screenshot.replace(/^data:image\/\w+;base64,/, ""),
+        instruction: params.instruction,
+        action_history: params.actionHistory
       }
     );
-    const url = buildUrl(this.appiumSessionConfig.serverUrl, `/session/${this.appiumSessionConfig.id}`);
-    await axios.delete(url);
-    console.log(">> Session stopped.");
-    this.gptDriverSessionId = void 0;
+    return {
+      gptCommands: response.data.gpt_commands,
+      appetizeCommands: response.data.appetize_commands,
+      actionHistory: response.data.action_history
+    };
+  }
+  // ─────────────────────────────────────────────────────────────────────────────
+  // DEVICE ACTION METHODS
+  // ─────────────────────────────────────────────────────────────────────────────
+  async getWdioClient() {
+    if (!this.appiumSessionStarted) {
+      await this.startSession();
+    }
+    if (this.driver?.sessionId != null) {
+      return this.driver;
+    }
+    const url = this.appiumSessionConfig.serverUrl;
+    const parsed = new URL(url);
+    const client = await webdriverio.attach({
+      sessionId: this.appiumSessionConfig.id,
+      options: {
+        protocol: parsed.protocol.replace(":", ""),
+        hostname: parsed.hostname,
+        port: parsed.port ? Number(parsed.port) : parsed.protocol === "https:" ? 443 : 80,
+        path: parsed.pathname && parsed.pathname !== "/" ? parsed.pathname : "/"
+      }
+    });
+    this.driver = client;
+    return client;
+  }
+  /**
+   * Performs a tap action at the specified coordinates.
+   */
+  async performTap(x, y) {
+    const client = await this.getWdioClient();
+    await client.performActions([
+      {
+        type: "pointer",
+        id: "finger1",
+        parameters: { pointerType: "touch" },
+        actions: [
+          { type: "pointerMove", duration: 0, x, y },
+          { type: "pointerDown", button: 0 },
+          { type: "pause", duration: 100 },
+          { type: "pointerUp", button: 0 }
+        ]
+      }
+    ]);
+  }
+  async performType(text) {
+    const client = await this.getWdioClient();
+    await client.keys(text.split(""));
+  }
+  async performScroll(direction) {
+    const client = await this.getWdioClient();
+    const w = this.appiumSessionConfig?.size?.width ?? 1080;
+    const h = this.appiumSessionConfig?.size?.height ?? 1920;
+    const x = Math.round(w / 2);
+    const startY = direction === "down" ? Math.round(h * 0.8) : Math.round(h * 0.2);
+    const endY = direction === "down" ? Math.round(h * 0.2) : Math.round(h * 0.8);
+    await client.performActions([
+      {
+        type: "pointer",
+        id: "finger1",
+        parameters: { pointerType: "touch" },
+        actions: [
+          { type: "pointerMove", duration: 0, x, y: startY },
+          { type: "pointerDown", button: 0 },
+          { type: "pause", duration: 100 },
+          { type: "pointerMove", duration: 500, x, y: endY },
+          { type: "pointerUp", button: 0 }
+        ]
+      }
+    ]);
+  }
+  async getPageSource() {
+    const client = await this.getWdioClient();
+    return client.getPageSource();
+  }
+  async performScrollUntil(params) {
+    const { direction, text, elementId } = params;
+    const max = params.maxScrolls ?? 10;
+    for (let i = 0; i < max; i++) {
+      const source = await this.getPageSource();
+      const found = elementId ? source.includes(elementId) : text ? source.includes(text) : false;
+      if (found) {
+        return;
+      }
+      await this.performScroll(direction);
+      await this._delay(500);
+    }
+    throw new Error(`scrollUntil target not found after ${max} scroll(s)`);
+  }
+  async getScreenshot(appiumSessionConfig) {
+    globalLogger.debug("Capturing screenshot...");
+    const url = buildUrl(this.appiumSessionConfig.serverUrl, `/session/${this.appiumSessionConfig.id}/screenshot`);
+    const screenshotResponse = await axios.get(url);
+    let screenshot = await screenshotResponse.data.value;
+    if (appiumSessionConfig.platform === "iOS") {
+      globalLogger.debug(`Resizing iOS screenshot to ${appiumSessionConfig.size.width}x${appiumSessionConfig.size.height}`);
+      const imageBuffer = Buffer.from(screenshot, "base64");
+      const transformedImage = await sharp(imageBuffer).resize(appiumSessionConfig.size.width, appiumSessionConfig.size.height).toBuffer();
+      screenshot = transformedImage.toString("base64");
+    }
+    return screenshot;
+  }
+  /**
+   * Helper method to delay execution.
+   *
+   * @private
+   */
+  _delay(ms) {
+    return new Promise((resolve) => setTimeout(resolve, ms));
   }
+  // ─────────────────────────────────────────────────────────────────────────────
+  // PUBLIC API METHODS
+  // ─────────────────────────────────────────────────────────────────────────────
   /**
    * Executes a specified command within the WebDriver session, optionally using an Appium handler.
    *
@@ -185,6 +971,7 @@ class GptDriver {
    * the command-specific operations. After executing the handler, the executed commands get logged on the GPTDriver servers.
    * If the handler execution fails or no handler is provided, the command gets executed by the GPTDriver using just natural language.
    *
+   * @deprecated Use `aiExecute()` instead. This method will be removed in a future version.
    * @param {string} command - The natural language command to be executed by the GPTDriver.
    * @param {AppiumHandler} [appiumHandler] - An optional function that processes Appium-specific commands.
    *                                          If provided, this handler is executed instead of calling the GPTDriver serves.
@@ -192,24 +979,136 @@ class GptDriver {
    * @throws {Error} If an error occurs during the execution of the Appium handler or while processing the command by the GPTDriver.
    */
   async execute(command, appiumHandler) {
-    console.log(">> Executing command:", command);
+    globalLogger.warn("Method 'execute()' is deprecated. Please use 'aiExecute()' instead.");
+    if (!this.appiumSessionStarted) {
+      await this.startSession();
+    }
+    globalLogger.info(`Executing command: ${command}`);
     const driver = this.driver;
     if (appiumHandler != null) {
       try {
+        await this.takeScreenshotAndLogCodeExecution(appiumHandler.toString());
         await appiumHandler(driver);
-        const screenshot = await this.getScreenshot(this.appiumSessionConfig);
-        await axios.post(`${this.gptDriverBaseUrl}/sessions/${this.gptDriverSessionId}/log_code_execution`, {
-          api_key: this.apiKey,
-          base64_screenshot: screenshot,
-          command: appiumHandler.toString()
-        });
+        globalLogger.debug("Custom Appium handler executed successfully");
       } catch (e) {
+        globalLogger.warn("Custom Appium handler failed, falling back to GPT handler");
         await this.gptHandler(command);
       }
     } else {
       await this.gptHandler(command);
     }
   }
+  /**
+   * Executes a specified command within the WebDriver session with configurable caching options.
+   *
+   * This is the recommended method for executing commands. It provides fine-grained control over
+   * caching behavior, allowing you to optimize performance and costs for repetitive test scenarios.
+   *
+   * If an `appiumHandler` is provided, it will be invoked with the WebDriver instance to perform
+   * the command-specific operations. After executing the handler, the executed commands get logged
+   * on the GPTDriver servers. If the handler execution fails or no handler is provided, the command
+   * gets executed by the GPTDriver using natural language processing.
+   *
+   * @param {Object} params - The execution parameters
+   * @param {string} params.command - The natural language command to be executed by the GPTDriver.
+   *                                  Examples: "Click the login button", "Enter 'test@example.com' in the email field"
+   * @param {AppiumHandler} [params.appiumHandler] - An optional function that processes Appium-specific commands.
+   *                                                  If provided, this handler is executed instead of calling
+   *                                                  the GPTDriver API. Useful for performance optimization when
+   *                                                  you know the exact Appium commands to execute.
+   * @param {CachingMode} [params.cachingMode] - Controls how the GPTDriver caches this command execution.
+   *                                             If not specified, uses the global caching mode set in the constructor.
+   *                                             Options:
+   *                                             - "NONE"
+   *                                             - "FULL_SCREEN"
+   *                                             - "INTERACTION_REGION"
+   * @param {boolean} [params.useSmartLoop] - If true, uses the smart loop execution (Cache -> AI -> Execute -> Populate)
+   *                                          which optimizes execution by checking cache first and populating it after.
+   *                                          Default: false (uses legacy gptHandler)
+   *
+   * @returns {Promise<void>} A promise that resolves when the command execution is complete.
+   *
+   * @throws {Error} If an error occurs during the execution of the Appium handler or while processing
+   *                 the command by the GPTDriver.
+   *
+   * @example
+   * // Basic usage with natural language (no caching)
+   * await driver.aiExecute({
+   *   command: "Click the submit button"
+   * });
+   *
+   * @example
+   * // Full screen caching for repetitive navigation on similar screens
+   * await driver.aiExecute({
+   *   command: "Navigate to the settings page",
+   *   cachingMode: "FULL_SCREEN"
+   * });
+   *
+   * @example
+   * // Interaction region caching for repeated actions on the same button
+   * await driver.aiExecute({
+   *   command: "Click the login button",
+   *   cachingMode: "INTERACTION_REGION"
+   * });
+   *
+   * @example
+   * // With custom Appium handler as fallback
+   * await driver.aiExecute({
+   *   command: "Click the login button",
+   *   appiumHandler: async (driver) => {
+   *     const loginBtn = await driver.$('~loginButton');
+   *     await loginBtn.click();
+   *   },
+   *   cachingMode: "INTERACTION_REGION"
+   * });
+   *
+   * @example
+   * // Force fresh execution for dynamic content
+   * await driver.aiExecute({
+   *   command: "Verify the current timestamp",
+   *   cachingMode: "NONE"
+   * });
+   *
+   * @example
+   * // Using smart loop for optimized caching
+   * await driver.aiExecute({
+   *   command: "Click the login button",
+   *   useSmartLoop: true,
+   *   cachingMode: "FULL_SCREEN"
+   * });
+   */
+  async aiExecute({ command, appiumHandler, cachingMode, useSmartLoop = false }) {
+    if (!this.appiumSessionStarted) {
+      await this.startSession();
+    }
+    globalLogger.info(`Executing command: ${command}`);
+    const driver = this.driver;
+    if (appiumHandler != null) {
+      try {
+        await this.takeScreenshotAndLogCodeExecution(appiumHandler.toString());
+        await appiumHandler(driver);
+        globalLogger.debug("Custom Appium handler executed successfully");
+        this.step_number++;
+        return;
+      } catch (e) {
+        globalLogger.warn("Custom Appium handler failed, falling back to AI execution");
+      }
+    }
+    if (useSmartLoop) {
+      const ctx = this.createSmartLoopContext();
+      const result = await executeSmartLoop(ctx, {
+        stepNumber: this.step_number,
+        description: command,
+        instruction: command
+      });
+      if (!result.success) {
+        throw new Error(result.error || "Smart loop execution failed");
+      }
+      this.step_number++;
+    } else {
+      await this.gptHandler(command, cachingMode);
+    }
+  }
   /**
    * Asserts a single condition using the GPTDriver.
    *
@@ -217,13 +1116,25 @@ class GptDriver {
    * If the assertion fails, an error is thrown.
    *
    * @param {string} assertion - The condition to be asserted.
+   * @param cachingMode - The caching mode to be used for the assertion.
    * @throws {Error} If the assertion fails.
    */
-  async assert(assertion) {
-    console.log(">> Asserting:", assertion);
-    const results = await this.checkBulk([assertion]);
-    if (!Object.values(results).at(0)) {
-      throw new Error(`Failed assertion: ${assertion}`);
+  async assert(assertion, cachingMode) {
+    if (!this.appiumSessionStarted) {
+      await this.startSession();
+    }
+    try {
+      const results = await this.checkBulk([assertion], cachingMode);
+      if (!Object.values(results).at(0)) {
+        await this.setSessionStatus("failed");
+        globalLogger.error(`Assertion failed: ${assertion}`);
+        throw new Error(`Failed assertion: ${assertion}`);
+      }
+      this.step_number = this.step_number + 1;
+      globalLogger.info(`Assertion passed: ${assertion}`);
+    } catch (e) {
+      await this.setSessionStatus("failed");
+      throw e;
     }
   }
   /**
@@ -233,43 +1144,100 @@ class GptDriver {
    * If any assertion fails, an error is thrown listing all failed assertions.
    *
    * @param {string[]} assertions - An array of conditions to be asserted.
+   * @param cachingMode - The caching mode to be used for the assertions.
    * @throws {Error} If any of the assertions fail.
    */
-  async assertBulk(assertions) {
-    console.log(">> Asserting:", assertions);
-    const results = await this.checkBulk(assertions);
-    const failedAssertions = Object.values(results).reduce((prev, current, currentIndex) => {
-      if (!current) {
-        return [...prev, assertions.at(currentIndex)];
+  async assertBulk(assertions, cachingMode) {
+    if (!this.appiumSessionStarted) {
+      await this.startSession();
+    }
+    try {
+      const results = await this.checkBulk(assertions, cachingMode);
+      const failedAssertions = Object.values(results).reduce((prev, current, currentIndex) => {
+        if (!current) {
+          return [...prev, assertions.at(currentIndex)];
+        }
+        return prev;
+      }, []);
+      if (failedAssertions.length > 0) {
+        await this.setSessionStatus("failed");
+        globalLogger.error(`Multiple assertions failed: ${failedAssertions.join(", ")}`);
+        throw new Error(`Failed assertions: ${failedAssertions.join(", ")}`);
       }
-      return prev;
-    }, []);
-    if (failedAssertions.length > 0) {
-      throw new Error(`Failed assertions: ${failedAssertions.join(", ")}`);
+      this.step_number = this.step_number + 1;
+      globalLogger.info(`All ${assertions.length} assertions passed`);
+    } catch (e) {
+      await this.setSessionStatus("failed");
+      throw e;
     }
   }
   /**
    * Checks multiple conditions and returns their results using the GPTDriver.
    *
    * This method sends a bulk condition request and returns the results of the conditions.
+   * Failed conditions will be retried up to maxRetries times.
    *
    * @param {string[]} conditions - An array of conditions to be checked.
+   * @param {CachingMode} cachingMode - The caching mode to be used for the conditions.
+   * @param {number} maxRetries - The maximum number of retries if any condition fails (default: 2).
+   * @param {number} retryDelayMs - The delay in milliseconds between retries (default: 1000).
    * @returns {Promise<Record<string, boolean>>} A promise that resolves with an object mapping each condition
    *                                             to a boolean indicating whether the condition was met.
    */
-  async checkBulk(conditions) {
-    console.log(">> Checking:", conditions);
-    const screenshot = await this.getScreenshot(this.appiumSessionConfig);
-    const response = await axios.post(
-      `${this.gptDriverBaseUrl}/sessions/${this.gptDriverSessionId}/assert`,
-      {
-        api_key: this.apiKey,
-        base64_screenshot: screenshot,
-        assertions: conditions,
-        command: `Assert: ${JSON.stringify(conditions)}`
+  async checkBulk(conditions, cachingMode, maxRetries = 2, retryDelayMs = 1e3) {
+    let attempt = 0;
+    let results = {};
+    while (attempt <= maxRetries) {
+      results = await this._checkBulkOnce(conditions, cachingMode, attempt);
+      const failedConditions = Object.entries(results).filter(([_, success]) => !success).map(([key, _]) => key);
+      if (failedConditions.length === 0) {
+        return results;
       }
-    );
-    return response.data.results;
+      attempt++;
+      if (attempt <= maxRetries) {
+        globalLogger.info(
+          `>>	 Conditions failed ${JSON.stringify(failedConditions)}. Retrying in ${retryDelayMs}ms... (Attempt ${attempt}/${maxRetries})`
+        );
+        await this._delay(retryDelayMs);
+      } else {
+        globalLogger.info(`>>	 Conditions failed: ${JSON.stringify(failedConditions)}`);
+      }
+    }
+    return results;
+  }
+  /**
+   * Internal method to check conditions once without retry logic.
+   *
+   * @private
+   */
+  async _checkBulkOnce(conditions, cachingMode, attempt = 0) {
+    if (!this.appiumSessionStarted) {
+      await this.startSession();
+    }
+    globalLogger.info(`Checking conditions (attempt ${attempt}): ${conditions.join(", ")}`);
+    try {
+      let screenshot;
+      if (!this.useGptDriverCloud) {
+        screenshot = await this.getScreenshot(this.appiumSessionConfig);
+      }
+      const response = await axios.post(
+        `${this.gptDriverBaseUrl}/sessions/${this.gptDriverSessionId}/assert`,
+        {
+          api_key: this.apiKey,
+          base64_screenshot: screenshot,
+          assertions: conditions,
+          command: `Assert: ${JSON.stringify(conditions)}`,
+          caching_mode: cachingMode ?? this.cachingMode,
+          step_number: this.step_number
+        }
+      );
+      globalLogger.debug(`Check results: ${JSON.stringify(response.data.results)}`);
+      return response.data.results;
+    } catch (e) {
+      globalLogger.error("Failed to check conditions", e);
+      await this.setSessionStatus("failed");
+      throw e;
+    }
   }
   /**
    * Extracts specified information using the GPTDriver.
@@ -279,30 +1247,305 @@ class GptDriver {
    *
    * @param {string[]} extractions - An array of extraction criteria. Each criterion specifies what information
    *                                 should be extracted from the session.
+   * @param cachingMode - The caching mode to be used for the extraction.
    * @returns {Promise<Record<string, any>>} A promise that resolves with an object mapping each extraction criterion
    *                                         to the extracted data. The structure of the returned data depends on the
    *                                         specifics of the extraction criteria.
    */
-  async extract(extractions) {
-    console.log(">> Extracting:", extractions);
-    const screenshot = await this.getScreenshot(this.appiumSessionConfig);
+  async extract(extractions, cachingMode) {
+    if (!this.appiumSessionStarted) {
+      await this.startSession();
+    }
+    globalLogger.info(`Extracting data: ${extractions.join(", ")}`);
+    let screenshot;
+    if (!this.useGptDriverCloud) {
+      screenshot = await this.getScreenshot(this.appiumSessionConfig);
+    }
     const response = await axios.post(
       `${this.gptDriverBaseUrl}/sessions/${this.gptDriverSessionId}/extract`,
       {
         api_key: this.apiKey,
         base64_screenshot: screenshot,
         extractions,
-        command: `Extract: ${JSON.stringify(extractions)}`
+        command: `Extract: ${JSON.stringify(extractions)}`,
+        step_number: this.step_number
       }
     );
+    this.step_number = this.step_number + 1;
+    globalLogger.debug(`Extraction results: ${JSON.stringify(response.data.results)}`);
     return response.data.results;
   }
-  async gptHandler(command) {
+  /**
+   * Opens a deep link url in the Appium session.
+   *
+   * This method sends a request to the GPT Driver server to open a deep link url in the Appium session.
+   *
+   * @param {OpenDeepLinkUrlParams} params - The parameters for opening the deep link url.
+   * @returns {Promise<void>} A promise that resolves when the deep link url is opened.
+   */
+  async openDeepLinkUrl(params) {
+    if (!this.appiumSessionStarted) {
+      await this.startSession();
+    }
+    globalLogger.info(`Opening deep link: ${params.url}`);
+    if (params.package == null && this.appiumSessionConfig?.platform === "Android") {
+      throw new Error("Package is required for Android platform");
+    }
+    await this.executeCommand(
+      {
+        url: `http://localhost:4723/session/${this.appiumSessionConfig?.id}/execute/sync`,
+        method: "POST",
+        data: {
+          "script": "mobile:deepLink",
+          "args": [{
+            url: params.url,
+            ...params.bundleId && { bundleId: params.bundleId },
+            ...params.package && { package: params.package }
+          }]
+        }
+      }
+    );
+    this.step_number = this.step_number + 1;
+    globalLogger.debug("Deep link opened successfully");
+  }
+  /**
+   * Reads a flow JSON file from disk and validates it using the SavableTestStoreSchema.
+   *
+   * Returns the parsed and validated object on success; throws a detailed error on failure.
+   *
+   * @param filePath - Path to the flow file (JSON)
+   * @param options - Optional execution options
+   * @param options.useSmartLoop - If true, uses the smart loop execution (Cache -> AI -> Execute -> Populate)
+   *                               for AI, tap, and assert steps. This optimizes execution by checking cache
+   *                               first and populating it after successful execution. Default: false
+   * @returns The validated flow data
+   *
+   * @example
+   * // Execute flow with default settings (legacy gptHandler)
+   * const result = await driver.executeFlow('tests/login-flow.json');
+   *
+   * @example
+   * // Execute flow with smart loop enabled for optimized caching
+   * const result = await driver.executeFlow('tests/login-flow.json', { useSmartLoop: true });
+   */
+  async executeFlow(filePath, options) {
+    const useSmartLoop = options?.useSmartLoop ?? false;
+    globalLogger.info(`Loading flow from file: ${filePath}`);
+    const absolutePath = path.resolve(filePath);
+    const baseDir = path.dirname(absolutePath);
+    let raw;
+    try {
+      raw = await node_fs.promises.readFile(absolutePath, "utf-8");
+    } catch (e) {
+      const msg = `Failed to read file at ${filePath}: ${e?.message ?? e}`;
+      globalLogger.error(msg);
+      throw new Error(msg);
+    }
+    let json;
+    try {
+      json = JSON.parse(raw);
+    } catch (e) {
+      const msg = `Invalid JSON in flow file ${filePath}: ${e?.message ?? e}`;
+      globalLogger.error(msg);
+      throw new Error(msg);
+    }
+    const parsed = SavableTestStoreSchema.safeParse(json);
+    if (!parsed.success) {
+      const issues = parsed.error.issues.map((iss) => `- ${iss.path.join(".") || "<root>"}: ${iss.message}`).join("\n");
+      const msg = `Flow validation failed for ${filePath}:
+${issues}`;
+      globalLogger.error(msg);
+      throw new Error(msg);
+    }
+    const rootFlow = parsed.data;
+    globalLogger.info(`Flow file validated successfully: ${filePath}`);
+    const visited = /* @__PURE__ */ new Set();
+    const loadFlow = async (p) => {
+      const abs = path.isAbsolute(p) ? p : path.resolve(baseDir, p);
+      const rawChild = await node_fs.promises.readFile(abs, "utf-8");
+      const childJson = JSON.parse(rawChild);
+      const val = SavableTestStoreSchema.safeParse(childJson);
+      if (!val.success) {
+        const issues = val.error.issues.map((iss) => `- ${iss.path.join(".") || "<root>"}: ${iss.message}`).join("\n");
+        throw new Error(`Flow validation failed for referenced file ${abs}:
+${issues}`);
+      }
+      return val.data;
+    };
+    const expandSteps = async (steps, inheritedParams, parentDir, stack) => {
+      const out = [];
+      for (const step of steps) {
+        if (step.type === "fileRef") {
+          const refPath = path.isAbsolute(step.path) ? step.path : path.resolve(parentDir, step.path);
+          const refKey = path.normalize(refPath);
+          if (visited.has(refKey)) {
+            const cycle = [...stack, refKey].map((p) => path.basename(p)).join(" -> ");
+            throw new Error(`Detected circular fileRef: ${cycle}`);
+          }
+          visited.add(refKey);
+          const child = await loadFlow(refPath);
+          const mergedParams = { ...inheritedParams, ...step.overrides ?? {} };
+          const childDir = path.dirname(refPath);
+          const childExpanded = await expandSteps(child.steps, mergedParams, childDir, [...stack, refKey]);
+          out.push(...childExpanded);
+        } else {
+          const resolved = { ...step, __params: { ...inheritedParams } };
+          out.push(resolved);
+        }
+      }
+      return out;
+    };
+    const effectiveParams = { ...rootFlow.params ?? {} };
+    const expandedSteps = await expandSteps(rootFlow.steps, effectiveParams, baseDir, [absolutePath]);
+    if (!this.appiumSessionStarted) {
+      await this.startSession();
+    }
+    globalLogger.info(`Executing flow '${rootFlow.name}' with ${expandedSteps.length} step(s)...`);
+    let executed = 0;
+    try {
+      for (const step of expandedSteps) {
+        const params = step.__params ?? effectiveParams;
+        const prefix = `Step #${executed + 1} [${step.type}${step.optional ? ", optional" : ""}]`;
+        try {
+          switch (step.type) {
+            case "ai": {
+              const instruction = this.interpolateTemplate(step.instruction, params);
+              globalLogger.info(`${prefix}: ${instruction}`);
+              if (useSmartLoop) {
+                const ctx = this.createSmartLoopContext();
+                const result = await executeSmartLoop(ctx, {
+                  stepNumber: this.step_number,
+                  description: instruction,
+                  instruction
+                });
+                if (!result.success) {
+                  throw new Error(result.error || "Smart loop execution failed");
+                }
+                this.step_number++;
+              } else {
+                await this.aiExecute({ command: instruction });
+              }
+              break;
+            }
+            case "tap": {
+              const description = step.descriptionText ? this.interpolateTemplate(step.descriptionText, params) : void 0;
+              if (!description) {
+                throw new Error("Tap step requires a descriptionText. Coordinate-based taps are no longer supported.");
+              }
+              globalLogger.info(`${prefix}: ${description}`);
+              if (useSmartLoop) {
+                const ctx = this.createSmartLoopContext();
+                const result = await executeSmartLoop(ctx, {
+                  stepNumber: this.step_number,
+                  description,
+                  instruction: description
+                });
+                if (!result.success) {
+                  throw new Error(result.error || "Smart loop execution failed");
+                }
+                this.step_number++;
+              } else {
+                await this.aiExecute({ command: description });
+              }
+              break;
+            }
+            case "assert": {
+              const description = step.descriptionText ? this.interpolateTemplate(step.descriptionText, params) : void 0;
+              if (!description) {
+                throw new Error("Assert step requires a descriptionText. Coordinate-based assertions are no longer supported.");
+              }
+              globalLogger.info(`${prefix}: ${description}`);
+              if (useSmartLoop) {
+                const instruction = `Verify that: ${description}`;
+                const ctx = this.createSmartLoopContext();
+                const result = await executeSmartLoop(ctx, {
+                  stepNumber: this.step_number,
+                  description,
+                  instruction
+                });
+                if (!result.success) {
+                  throw new Error(result.error || "Smart loop execution failed");
+                }
+                this.step_number++;
+              } else {
+                await this.assert(description);
+              }
+              break;
+            }
+            case "type": {
+              const text = this.interpolateTemplate(step.text, params);
+              globalLogger.info(`${prefix}: Type text`);
+              await this.takeScreenshotAndLogCodeExecution(`type: text=${text}`);
+              await this.performType(text);
+              this.step_number++;
+              break;
+            }
+            case "scroll": {
+              globalLogger.info(`${prefix}: Scroll ${step.direction}`);
+              await this.takeScreenshotAndLogCodeExecution(`scroll: direction=${step.direction}`);
+              await this.performScroll(step.direction);
+              this.step_number++;
+              break;
+            }
+            case "zoom": {
+              globalLogger.info(`${prefix}: Zoom ${step.direction}`);
+              await this.takeScreenshotAndLogCodeExecution(`zoom: direction=${step.direction}`);
+              this.step_number++;
+              break;
+            }
+            case "scrollUntil": {
+              const interpolatedText = step.text != null ? this.interpolateTemplate(step.text, params) : void 0;
+              globalLogger.info(`${prefix}: Scroll until ${interpolatedText ?? step.elementId}`);
+              await this.takeScreenshotAndLogCodeExecution(`scrollUntil: text=${interpolatedText}, elementId=${step.elementId}`);
+              await this.performScrollUntil({
+                direction: step.direction,
+                text: interpolatedText,
+                elementId: step.elementId,
+                maxScrolls: step.maxScrolls
+              });
+              this.step_number++;
+              break;
+            }
+            case "deeplink": {
+              const pkg = params["package"];
+              const bundleId = params["bundleId"];
+              const url = this.interpolateTemplate(step.url, params);
+              globalLogger.info(`${prefix}: Open deeplink ${url}`);
+              await this.takeScreenshotAndLogCodeExecution(`openDeepLinkUrl: url=${url}`);
+              await this.openDeepLinkUrl({ url, package: pkg, bundleId });
+              break;
+            }
+            default: {
+              throw new Error(`Unsupported step type at execution: ${step.type}`);
+            }
+          }
+          executed++;
+        } catch (err) {
+          if (step.optional) {
+            globalLogger.warn(`${prefix} failed but marked optional. Continuing. Error: ${err.message}`);
+            continue;
+          }
+          throw err;
+        }
+      }
+    } catch (e) {
+      try {
+        await this.setSessionStatus("failed");
+      } catch {
+      }
+      throw e;
+    }
+    return rootFlow;
+  }
+  async gptHandler(command, cachingMode) {
     try {
       let conditionSucceeded = false;
       while (!conditionSucceeded) {
-        const screenshot = await this.getScreenshot(this.appiumSessionConfig);
-        console.log(">> Asking GTP Driver for next action...");
+        let screenshot;
+        if (!this.useGptDriverCloud) {
+          screenshot = await this.getScreenshot(this.appiumSessionConfig);
+        }
+        globalLogger.info("Requesting next action from GPT Driver...");
         const response = await axios.request(
           {
             url: `${this.gptDriverBaseUrl}/sessions/${this.gptDriverSessionId}/execute`,
@@ -310,39 +1553,49 @@ class GptDriver {
             data: {
               api_key: this.apiKey,
               command,
-              base64_screenshot: screenshot
+              base64_screenshot: screenshot,
+              caching_mode: cachingMode ?? this.cachingMode,
+              step_number: this.step_number
             }
           }
         );
         const executeStatus = response.data.status;
         if (executeStatus === "failed") {
-          const errorMessage = response?.data?.commands?.at(0)?.data;
+          const errorMessage = response.data?.commands?.at(0)?.data;
+          globalLogger.error(`Execution failed: ${errorMessage ?? "Unknown error"}`);
           throw new Error(errorMessage ?? "Execution failed");
         }
         conditionSucceeded = executeStatus !== "inProgress";
         const executeResponse = response.data;
-        for (const command2 of executeResponse.commands) {
-          await this.executeCommand(command2);
+        globalLogger.debug(`Received ${executeResponse.commands.length} command(s) to execute`);
+        for (const appiumCommand of executeResponse.commands) {
+          await this.executeCommand(appiumCommand);
         }
         if (!conditionSucceeded) {
+          globalLogger.debug("Command still in progress, waiting...");
           await delay(1500);
         }
       }
+      this.step_number = this.step_number + 1;
+      globalLogger.info("Command execution completed successfully");
     } catch (e) {
-      await this.stopSession("failed");
+      globalLogger.error("GPT handler failed", e);
+      await this.setSessionStatus("failed");
       throw e;
     }
   }
   async executeCommand(command) {
-    const firstAction = command.data.actions?.at(0);
+    const firstAction = command.data?.actions?.at(0);
     if (firstAction?.type === "pause" && firstAction.duration != null) {
+      globalLogger.debug(`Pausing for ${firstAction.duration} seconds`);
       await delay(firstAction * 1e3);
-    } else {
+    } else if (!this.useGptDriverCloud) {
       const parsedUrl = new URL(command.url);
       parsedUrl.protocol = this.appiumSessionConfig.serverUrl.protocol;
       parsedUrl.host = this.appiumSessionConfig.serverUrl.host;
       parsedUrl.port = this.appiumSessionConfig.serverUrl.port != "" ? `${this.appiumSessionConfig.serverUrl.port}` : "";
       parsedUrl.pathname = this.appiumSessionConfig.serverUrl.pathname != "/" ? `${this.appiumSessionConfig.serverUrl.pathname}${parsedUrl.pathname}` : parsedUrl.pathname;
+      globalLogger.debug(`Executing ${command.method} request to ${parsedUrl.pathname}`);
       await axios.request({
         url: parsedUrl.toString(),
         method: command.method,
@@ -350,16 +1603,25 @@ class GptDriver {
       });
     }
   }
-  async getScreenshot(appiumSessionConfig) {
-    const url = buildUrl(this.appiumSessionConfig.serverUrl, `/session/${this.appiumSessionConfig.id}/screenshot`);
-    const screenshotResponse = await axios.get(url);
-    let screenshot = await screenshotResponse.data.value;
-    if (appiumSessionConfig.platform === "iOS") {
-      const imageBuffer = Buffer.from(screenshot, "base64");
-      const transformedImage = await sharp(imageBuffer).resize(appiumSessionConfig.size.width, appiumSessionConfig.size.height).toBuffer();
-      screenshot = transformedImage.toString("base64");
+  async logCodeExecution(screenshot, command) {
+    try {
+      const screenshot2 = await this.getScreenshot(this.appiumSessionConfig);
+      await axios.post(`${this.gptDriverBaseUrl}/sessions/${this.gptDriverSessionId}/log_code_execution`, {
+        api_key: this.apiKey,
+        base64_screenshot: screenshot2,
+        command
+      });
+    } catch (e) {
+      globalLogger.error("Failed to log code execution", e);
+    }
+  }
+  async takeScreenshotAndLogCodeExecution(command) {
+    try {
+      const screenshot = await this.getScreenshot(this.appiumSessionConfig);
+      await this.logCodeExecution(screenshot, command);
+    } catch (e) {
+      globalLogger.error("Failed to log code execution", e);
     }
-    return screenshot;
   }
 }