sunholo 0.73.3__tar.gz → 0.74.0__tar.gz

{sunholo-0.73.3 → sunholo-0.74.0}/PKG-INFO

@@ -1,9 +1,9 @@
 Metadata-Version: 2.1
 Name: sunholo
-Version: 0.73.3
+Version: 0.74.0
 Summary: Large Language Model DevOps - a package to help deploy LLMs to the Cloud.
 Home-page: https://github.com/sunholo-data/sunholo-py
-Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.73.3.tar.gz
+Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.74.0.tar.gz
 Author: Holosun ApS
 Author-email: multivac@sunholo.com
 License: Apache License, Version 2.0
@@ -58,6 +58,7 @@ Requires-Dist: pg8000; extra == "all"
 Requires-Dist: pgvector; extra == "all"
 Requires-Dist: pillow; extra == "all"
 Requires-Dist: playwright; extra == "all"
+Requires-Dist: psutil; extra == "all"
 Requires-Dist: psycopg2-binary; extra == "all"
 Requires-Dist: pypdf; extra == "all"
 Requires-Dist: python-socketio; extra == "all"
@@ -67,7 +68,7 @@ Requires-Dist: supabase; extra == "all"
 Requires-Dist: tabulate; extra == "all"
 Requires-Dist: tantivy; extra == "all"
 Requires-Dist: tiktoken; extra == "all"
-Requires-Dist: unstructured[local-inference]; extra == "all"
+Requires-Dist: unstructured[local-inference]==0.14.9; extra == "all"
 Provides-Extra: cli
 Requires-Dist: jsonschema>=4.21.1; extra == "cli"
 Requires-Dist: rich; extra == "cli"
@@ -83,10 +84,11 @@ Requires-Dist: tantivy; extra == "database"
 Provides-Extra: pipeline
 Requires-Dist: GitPython; extra == "pipeline"
 Requires-Dist: lark; extra == "pipeline"
+Requires-Dist: psutil; extra == "pipeline"
 Requires-Dist: pypdf; extra == "pipeline"
 Requires-Dist: pytesseract; extra == "pipeline"
 Requires-Dist: tabulate; extra == "pipeline"
-Requires-Dist: unstructured[local-inference]; extra == "pipeline"
+Requires-Dist: unstructured[local-inference]==0.14.9; extra == "pipeline"
 Provides-Extra: gcp
 Requires-Dist: google-api-python-client; extra == "gcp"
 Requires-Dist: google-cloud-alloydb-connector[pg8000]; extra == "gcp"

{sunholo-0.73.3 → sunholo-0.74.0}/setup.py

@@ -1,7 +1,7 @@
 from setuptools import setup, find_packages
 
 # Define your base version
-version = '0.73.3'
+version = '0.74.0'
 
 setup(
     name='sunholo',
@@ -73,6 +73,7 @@ setup(
             "pgvector",
             "pillow",
             "playwright",
+            "psutil",
             "psycopg2-binary",
             "pypdf",
             "python-socketio",
@@ -82,7 +83,7 @@ setup(
             "tabulate",
             "tantivy",
             "tiktoken",
-            "unstructured[local-inference]",
+            "unstructured[local-inference]==0.14.9",
 
         ],
         'cli': [
@@ -102,10 +103,11 @@ setup(
         'pipeline': [
             "GitPython",
             "lark",
+            "psutil",
             "pypdf",
             "pytesseract",
             "tabulate",
-            "unstructured[local-inference]",
+            "unstructured[local-inference]==0.14.9",
         ],
         'gcp': [
             "google-api-python-client",

{sunholo-0.73.3 → sunholo-0.74.0}/sunholo/cli/chat_vac.py

@@ -183,6 +183,10 @@ def stream_chat_session(service_url, service_name, stream=True):
             read_file = None
             read_file_count = None
             continue
+
+        if user_input.lower().startswith("!"):
+            console.print("[bold red]Could find no valid chat command for you, sorry[/bold red]")
+            continue
         
         if read_file:
             user_input = f"<user added file>{read_file}</user added file>\n{user_input}"

sunholo-0.74.0/sunholo/tools/__init__.py

@@ -0,0 +1 @@
+from .web_browser import BrowseWebWithImagePromptsBot

sunholo-0.74.0/sunholo/tools/web_browser.py

@@ -0,0 +1,355 @@
+import os
+import base64
+import json
+from datetime import datetime
+import urllib.parse
+
+from ..logging import log
+
+from ..utils.parsers import get_clean_website_name
+
+class BrowseWebWithImagePromptsBot:
+    """
+    BrowseWebWithImagePromptsBot is a base class for creating bots that interact with web pages using Playwright.
+    The bot can perform actions such as navigating, clicking, scrolling, typing text, and taking screenshots.
+    It also supports cookie management to maintain session state across interactions.
+
+    Methods:
+    - __init__(session_id, website_name, browser_type='chromium', headless=True):
+        Initializes the bot with the given session ID, website name, browser type, and headless mode.
+        Supported browser types: 'chromium', 'firefox', 'webkit'.
+
+    - load_cookies():
+        Loads cookies from a file and adds them to the browser context.
+
+    - save_cookies():
+        Saves the current cookies to a file.
+
+    - navigate(url):
+        Navigates to the specified URL.
+
+    - click(selector):
+        Clicks on the element specified by the selector.
+
+    - scroll(direction='down', amount=1):
+        Scrolls the page in the specified direction ('down', 'up', 'left', 'right') by the specified amount.
+
+    - type_text(selector, text):
+        Types the specified text into the element specified by the selector.
+
+    - take_screenshot():
+        Takes a screenshot and saves it with a timestamp in the session-specific directory. Returns the path to the screenshot.
+
+    - get_latest_screenshot_path():
+        Retrieves the path to the most recent screenshot in the session-specific directory.
+
+    - create_prompt_vars(current_action_description, session_goal):
+        Creates a dictionary of prompt variables from the current action description and session goal.
+
+    - send_screenshot_to_llm(screenshot_path, current_action_description="", session_goal=""):
+        Encodes the screenshot in base64, creates prompt variables, and sends them to the LLM. Returns the new instructions from the LLM.
+
+    - send_prompt_to_llm(prompt_vars, screenshot_base64):
+        Abstract method to be implemented by subclasses. Sends the prompt variables and screenshot to the LLM and returns the response.
+
+    - close():
+        Saves cookies, closes the browser, and stops Playwright.
+
+    - execute_instructions(instructions):
+        Executes the given set of instructions, takes a screenshot after each step, and sends the screenshot to the LLM for further instructions.
+
+    Example usage:
+
+    ```python
+    class ProductionBot(BrowseWebWithImagePromptsBot):
+        def send_prompt_to_llm(self, prompt_vars, screenshot_base64):
+            # Implement the actual logic to send the prompt and screenshot to the LLM and return the response
+            api_url = "https://api.example.com/process"  # Replace with the actual LLM API endpoint
+            headers = {"Content-Type": "application/json"}
+            data = {
+                "prompt": prompt_vars,
+                "screenshot": screenshot_base64
+            }
+            response = requests.post(api_url, headers=headers, data=json.dumps(data))
+            return response.text  # Assuming the response is in JSON format
+
+    @app.route('/run-bot', methods=['POST'])
+    def run_bot():
+        data = request.json
+        session_id = data.get('session_id')
+        website_name = data.get('website_name')
+        browser_type = data.get('browser_type', 'chromium')
+        current_action_description = data.get('current_action_description', "")
+        session_goal = data.get('session_goal', "")
+        
+        bot = ProductionBot(session_id=session_id, website_name=website_name, browser_type=browser_type, headless=True)
+        
+        # Check if initial instructions are provided
+        initial_instructions = data.get('instructions')
+        if initial_instructions:
+            bot.execute_instructions(initial_instructions)
+        else:
+            bot.execute_instructions([{'action':'navigate', 'url': website_name}])
+        
+        # Take initial screenshot and send to LLM
+        screenshot_path = bot.take_screenshot()
+        new_instructions = bot.send_screenshot_to_llm(screenshot_path, current_action_description, session_goal)
+        bot.execute_instructions(new_instructions)
+        
+        # Take final screenshot
+        bot.take_screenshot()
+        
+        bot.close()
+        
+        return jsonify({"status": "completed", "new_instructions": new_instructions})
+
+    if __name__ == "__main__":
+        app.run(host='0.0.0.0', port=8080)
+    ```
+    """
+#class BrowseWebWithImagePromptsBot:
+    def __init__(self, session_id, website_name, browser_type='chromium', headless=True, max_steps=10):
+        try:
+            from playwright.sync_api import sync_playwright
+        except ImportError as err:
+            print(err)
+            sync_playwright = None
+        if not sync_playwright:
+            raise ImportError("playright needed for BrowseWebWithImagePromptsBot class - install via `pip install sunholo[tools]`")
+        self.session_id = session_id or datetime.now().strftime("%Y%m%d%H%M%S")
+        self.website_name = website_name
+        self.browser_type = browser_type
+        self.max_steps = max_steps
+        self.steps = 0
+        self.screenshot_dir = f"browser_tool/{get_clean_website_name(website_name)}/{session_id}"
+        os.makedirs(self.screenshot_dir, exist_ok=True)
+        self.cookie_file = os.path.join(self.screenshot_dir, "cookies.json")
+        self.playwright = sync_playwright().start()
+        
+        if browser_type == 'chromium':
+            self.browser = self.playwright.chromium.launch(headless=headless)
+        elif browser_type == 'firefox':
+            self.browser = self.playwright.firefox.launch(headless=headless)
+        elif browser_type == 'webkit':
+            self.browser = self.playwright.webkit.launch(headless=headless)
+        else:
+            raise ValueError(f"Unsupported browser type: {browser_type}")
+        
+        self.context = self.browser.new_context()
+        self.page = self.context.new_page()
+        self.load_cookies()
+        self.actions_log = []
+        self.session_goal = None
+        self.session_screenshots = []
+
+    def load_cookies(self):
+        if os.path.exists(self.cookie_file):
+            with open(self.cookie_file, 'r') as f:
+                cookies = json.load(f)
+                self.context.add_cookies(cookies)
+
+    def save_cookies(self):
+        cookies = self.context.cookies()
+        with open(self.cookie_file, 'w') as f:
+            json.dump(cookies, f)
+
+    def navigate(self, url):
+        try:
+            self.page.goto(url)
+            self.page.wait_for_load_state()
+            log.info(f'Navigated to {url}')
+            self.actions_log.append(f"Navigated to {url}")
+        except Exception as err:
+            log.warning(f"navigate failed with {str(err)}")
+            self.actions_log.append(f"Tried to navigate to {url} but got an error")
+
+
+    def click(self, selector):
+        try:
+            self.page.click(selector)
+            self.page.wait_for_load_state()
+            log.info(f"Clicked on element with selector {selector}")
+            self.actions_log.append(f"Clicked on element with selector {selector}")
+        except Exception as err:
+            log.warning(f"click failed with {str(err)}")
+            self.actions_log.append(f"Tried to click on element with selector {selector} but got an error")            
+
+    def scroll(self, direction='down', amount=1):
+        try:
+            for _ in range(amount):
+                if direction == 'down':
+                    self.page.evaluate("window.scrollBy(0, window.innerHeight)")
+                elif direction == 'up':
+                    self.page.evaluate("window.scrollBy(0, -window.innerHeight)")
+                elif direction == 'left':
+                    self.page.evaluate("window.scrollBy(-window.innerWidth, 0)")
+                elif direction == 'right':
+                    self.page.evaluate("window.scrollBy(window.innerWidth, 0)")
+                self.page.wait_for_timeout(500)
+                log.info(f"Scrolled {direction} by {amount} page heights")
+                self.actions_log.append(f"Scrolled {direction} by {amount} page heights")
+        except Exception as err:
+            log.warning(f"Scrolled failed with {str(err)}")
+            self.actions_log.append(f"Tried to scroll {direction} by {amount} page heights but got an error")
+
+
+    def type_text(self, selector, text):
+        try:
+            self.page.fill(selector, text)
+            self.page.wait_for_load_state()
+            log.info(f"Typed text '{text}' into element with selector {selector}")
+            self.actions_log.append(f"Typed text '{text}' into element with selector {selector}")
+        except Exception as err:
+            log.warning(f"Typed text failed with {str(err)}")
+            self.actions_log.append(f"Tried to type text '{text}' into element with selector {selector} but got an error")
+
+    def take_screenshot(self, final=False):
+        timestamp = datetime.now().strftime("%Y%m%d%H%M%S")
+        parsed_url = urllib.parse.urlparse({self.page.url})
+        url_path = parsed_url.path
+        if final:
+            screenshot_path = os.path.join(self.screenshot_dir, f"final/{timestamp}_{url_path}.png")
+        else:
+            screenshot_path = os.path.join(self.screenshot_dir, f"{timestamp}_{url_path}.png")
+        self.page.screenshot(path=screenshot_path)
+        log.info(f"Screenshot {self.page.url} taken and saved to {screenshot_path}")
+        #self.actions_log.append(f"Screenshot {self.page.url} taken and saved to {screenshot_path}")
+        self.session_screenshots.append(screenshot_path)
+
+        return screenshot_path
+
+    def get_latest_screenshot_path(self):
+        screenshots = sorted(
+            [f for f in os.listdir(self.screenshot_dir) if f.startswith('screenshot_')],
+            key=lambda x: os.path.getmtime(os.path.join(self.screenshot_dir, x)),
+            reverse=True
+        )
+        if screenshots:
+            return os.path.join(self.screenshot_dir, screenshots[0])
+        return None
+
+    def create_prompt_vars(self, last_message):
+        prompt = {
+            "last_actions": self.actions_log,
+            "session_goal": self.session_goal,
+            "last_message": last_message
+        }
+        return prompt
+    
+    def check_llm_response(self, response):
+        if isinstance(response, dict):
+            output = response
+        elif isinstance(response, str):
+            output = json.loads(response)
+
+        #TODO: more validation
+        log.info(f'Response: {output=}')
+
+        if 'status' not in output:
+            log.error(f'Response did not contain status')
+
+        if 'new_instructions' not in output:
+            log.warning(f'Response did not include new_instructions')
+        
+        if 'message' not in output:
+            log.warning(f'Response did not include message')
+
+        return output        
+
+    def send_screenshot_to_llm(self, screenshot_path, last_message):
+        with open(screenshot_path, "rb") as image_file:
+            encoded_image = base64.b64encode(image_file.read()).decode('utf-8')
+        
+        prompt_vars = self.create_prompt_vars(last_message)
+        response = self.send_prompt_to_llm(prompt_vars, encoded_image)  # Sending prompt and image separately
+        
+        return self.check_llm_response(response)
+
+    def send_prompt_to_llm(self, prompt_vars, screenshot_base64):
+        raise NotImplementedError("""
+This method should be implemented by subclasses: `def send_prompt_to_llm(self, prompt_vars, screenshot_base64)`")
+        prompt = {
+            "last_actions": self.action_log,
+            "session_goal": self.session_goal,
+        }
+""")
+
+    def close(self):
+        self.save_cookies()
+        self.browser.close()
+        self.playwright.stop()
+
+    def execute_instructions(self, instructions: list, last_message: str=None):
+        if not instructions: 
+            log.info("No instructions found, returning immediately")
+            return
+        
+        if self.steps >= self.max_steps:
+            log.warning(f"Reached the maximum number of steps: {self.max_steps}")
+            return
+        
+        if not isinstance(instructions, list):
+            log.error(f"{instructions} {type(instructions)}")
+        for instruction in instructions:
+            if not isinstance(instruction, dict):
+                log.error(f"{instruction} {type(instruction)}")
+            action = instruction['action']
+            if action == 'navigate':
+                self.navigate(instruction['url'])
+            elif action == 'click':
+                self.click(instruction['selector'])
+            elif action == 'scroll':
+                self.scroll(instruction.get('direction', 'down'), instruction.get('amount', 1))
+            elif action == 'type':
+                self.type_text(instruction['selector'], instruction['text'])
+            self.steps += 1
+            if self.steps >= self.max_steps:
+                log.warning(f"Reached the maximum number of steps: {self.max_steps}")
+                return
+            
+        screenshot_path = self.take_screenshot()
+        next_browser_instructions = self.send_screenshot_to_llm(
+                screenshot_path, 
+                last_message=last_message)
+            
+        return next_browser_instructions
+    
+    def start_session(self, instructions, session_goal):
+            self.session_goal = session_goal
+
+            if not instructions:
+                instructions = [{'action': 'navigate', 'url': self.website_name}]
+
+            next_instructions = self.execute_instructions(instructions)
+
+            in_session = True
+            while in_session:
+                if next_instructions and 'status' in next_instructions:
+                    if next_instructions['status'] == 'in-progress':
+                        log.info(f'Browser message: {next_instructions.get('message')}')
+                        if 'new_instructions' not in next_instructions:
+                            log.error('Browser status: "in-progress" but no new_instructions')
+                        last_message = next_instructions['message']
+                        log.info(f'Browser message: {last_message}')
+                        next_instructions = self.execute_instructions(next_instructions['new_instructions'], last_message=last_message)
+                    else:
+                        log.info(f'Session finished due to status={next_instructions["status"]}')
+                        in_session=False
+                        break
+                else:
+                    log.info('Session finished due to next_instructions being empty')
+                    in_session=False
+                    break
+            
+            log.info("Session finished")
+            final_path = self.take_screenshot(final=True)
+            self.close()
+            
+            return {
+                "website": self.website_name,
+                "log": self.actions_log,
+                "next_instructions": next_instructions,
+                "session_screenshots": self.session_screenshots,
+                "final_page": final_path,
+            }
+

{sunholo-0.73.3 → sunholo-0.74.0}/sunholo/utils/parsers.py

@@ -13,6 +13,7 @@
 #   limitations under the License.
 import re
 import hashlib
+import urllib.parse
 
 def validate_extension_id(ext_id):
     """
@@ -183,4 +184,8 @@ def escape_braces(text):
     # Replace single braces with double braces
     text = re.sub(r'(?<!{){(?!{)', '{{', text)  # Replace '{' with '{{' if not already double braced
     text = re.sub(r'(?<!})}(?!})', '}}', text)  # Replace '}' with '}}' if not already double braced
-    return text
+    return text
+
+def get_clean_website_name(url):
+    parsed_url = urllib.parse.urlparse(url)
+    return parsed_url.netloc

{sunholo-0.73.3 → sunholo-0.74.0}/sunholo.egg-info/PKG-INFO

@@ -1,9 +1,9 @@
 Metadata-Version: 2.1
 Name: sunholo
-Version: 0.73.3
+Version: 0.74.0
 Summary: Large Language Model DevOps - a package to help deploy LLMs to the Cloud.
 Home-page: https://github.com/sunholo-data/sunholo-py
-Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.73.3.tar.gz
+Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.74.0.tar.gz
 Author: Holosun ApS
 Author-email: multivac@sunholo.com
 License: Apache License, Version 2.0
@@ -58,6 +58,7 @@ Requires-Dist: pg8000; extra == "all"
 Requires-Dist: pgvector; extra == "all"
 Requires-Dist: pillow; extra == "all"
 Requires-Dist: playwright; extra == "all"
+Requires-Dist: psutil; extra == "all"
 Requires-Dist: psycopg2-binary; extra == "all"
 Requires-Dist: pypdf; extra == "all"
 Requires-Dist: python-socketio; extra == "all"
@@ -67,7 +68,7 @@ Requires-Dist: supabase; extra == "all"
 Requires-Dist: tabulate; extra == "all"
 Requires-Dist: tantivy; extra == "all"
 Requires-Dist: tiktoken; extra == "all"
-Requires-Dist: unstructured[local-inference]; extra == "all"
+Requires-Dist: unstructured[local-inference]==0.14.9; extra == "all"
 Provides-Extra: cli
 Requires-Dist: jsonschema>=4.21.1; extra == "cli"
 Requires-Dist: rich; extra == "cli"
@@ -83,10 +84,11 @@ Requires-Dist: tantivy; extra == "database"
 Provides-Extra: pipeline
 Requires-Dist: GitPython; extra == "pipeline"
 Requires-Dist: lark; extra == "pipeline"
+Requires-Dist: psutil; extra == "pipeline"
 Requires-Dist: pypdf; extra == "pipeline"
 Requires-Dist: pytesseract; extra == "pipeline"
 Requires-Dist: tabulate; extra == "pipeline"
-Requires-Dist: unstructured[local-inference]; extra == "pipeline"
+Requires-Dist: unstructured[local-inference]==0.14.9; extra == "pipeline"
 Provides-Extra: gcp
 Requires-Dist: google-api-python-client; extra == "gcp"
 Requires-Dist: google-cloud-alloydb-connector[pg8000]; extra == "gcp"

{sunholo-0.73.3 → sunholo-0.74.0}/sunholo.egg-info/requires.txt

@@ -39,6 +39,7 @@ pg8000
 pgvector
 pillow
 playwright
+psutil
 psycopg2-binary
 pypdf
 python-socketio
@@ -48,7 +49,7 @@ supabase
 tabulate
 tantivy
 tiktoken
-unstructured[local-inference]
+unstructured[local-inference]==0.14.9
 
 [anthropic]
 langchain-anthropic>=0.1.13
@@ -102,10 +103,11 @@ tiktoken
 [pipeline]
 GitPython
 lark
+psutil
 pypdf
 pytesseract
 tabulate
-unstructured[local-inference]
+unstructured[local-inference]==0.14.9
 
 [tools]
 openapi-spec-validator

sunholo-0.73.3/sunholo/tools/web_browser.py

@@ -1,169 +0,0 @@
-import os
-import base64
-import json
-from datetime import datetime
-try:
-    from playwright.sync_api import sync_playwright
-except ImportError:
-    sync_playwright = None
-
-class BrowseWebWithImagePromptsBot:
-    """
-    Examples:
-
-    ```python
-    class ProductionBot(BrowseWebWithImagePromptsBot):
-        def send_prompt_to_llm(self, prompt, screenshot_base64):
-            # Implement the actual logic to send the prompt and screenshot to the LLM and return the response
-            api_url = "https://api.example.com/process"  # Replace with the actual LLM API endpoint
-            headers = {"Content-Type": "application/json"}
-            data = {
-                "prompt": prompt,
-                "screenshot": screenshot_base64
-            }
-            response = requests.post(api_url, headers=headers, data=json.dumps(data))
-            return response.text  # Assuming the response is in JSON format
-
-    @app.route('/run-bot', methods=['POST'])
-    def run_bot():
-        data = request.json
-        session_id = data.get('session_id')
-        website_name = data.get('website_name')
-        browser_type = data.get('browser_type', 'chromium')
-        current_action_description = data.get('current_action_description', "")
-        next_goal = data.get('next_goal', "")
-        
-        bot = ProductionBot(session_id=session_id, website_name=website_name, browser_type=browser_type, headless=True)
-        
-        # Check if initial instructions are provided
-        initial_instructions = data.get('instructions')
-        if initial_instructions:
-            bot.execute_instructions(initial_instructions)
-        
-        # Take initial screenshot and send to LLM if no instructions provided
-        if not initial_instructions:
-            screenshot_path = bot.take_screenshot()
-            new_instructions = bot.send_screenshot_to_llm(screenshot_path, current_action_description, next_goal)
-            bot.execute_instructions(new_instructions)
-        
-        # Take final screenshot
-        bot.take_screenshot()
-        
-        bot.close()
-        
-        return jsonify({"status": "completed", "new_instructions": new_instructions})
-
-    if __name__ == "__main__":
-        app.run(host='0.0.0.0', port=8080)
-    ```
-    """
-    def __init__(self, session_id, website_name, browser_type='chromium', headless=True):
-        if not sync_playwright:
-            raise ImportError("playright needed for BrowseWebWithImagePromptsBot class - install via `pip install sunholo[tools]`")
-        self.session_id = session_id
-        self.website_name = website_name
-        self.browser_type = browser_type
-        self.screenshot_dir = f"{website_name}_{session_id}"
-        os.makedirs(self.screenshot_dir, exist_ok=True)
-        self.cookie_file = os.path.join(self.screenshot_dir, "cookies.json")
-        self.playwright = sync_playwright().start()
-        
-        if browser_type == 'chromium':
-            self.browser = self.playwright.chromium.launch(headless=headless)
-        elif browser_type == 'firefox':
-            self.browser = self.playwright.firefox.launch(headless=headless)
-        elif browser_type == 'webkit':
-            self.browser = self.playwright.webkit.launch(headless=headless)
-        else:
-            raise ValueError(f"Unsupported browser type: {browser_type}")
-        
-        self.context = self.browser.new_context()
-        self.page = self.context.new_page()
-        self.load_cookies()
-
-    def load_cookies(self):
-        if os.path.exists(self.cookie_file):
-            with open(self.cookie_file, 'r') as f:
-                cookies = json.load(f)
-                self.context.add_cookies(cookies)
-
-    def save_cookies(self):
-        cookies = self.context.cookies()
-        with open(self.cookie_file, 'w') as f:
-            json.dump(cookies, f)
-
-    def navigate(self, url):
-        self.page.goto(url)
-
-    def click(self, selector):
-        self.page.click(selector)
-
-    def scroll(self, direction='down', amount=1):
-        for _ in range(amount):
-            if direction == 'down':
-                self.page.evaluate("window.scrollBy(0, window.innerHeight)")
-            elif direction == 'up':
-                self.page.evaluate("window.scrollBy(0, -window.innerHeight)")
-            elif direction == 'left':
-                self.page.evaluate("window.scrollBy(-window.innerWidth, 0)")
-            elif direction == 'right':
-                self.page.evaluate("window.scrollBy(window.innerWidth, 0)")
-
-    def type_text(self, selector, text):
-        self.page.fill(selector, text)
-
-    def take_screenshot(self):
-        timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
-        screenshot_path = os.path.join(self.screenshot_dir, f"screenshot_{timestamp}.png")
-        self.page.screenshot(path=screenshot_path)
-        return screenshot_path
-
-    def get_latest_screenshot_path(self):
-        screenshots = sorted(
-            [f for f in os.listdir(self.screenshot_dir) if f.startswith('screenshot_')],
-            key=lambda x: os.path.getmtime(os.path.join(self.screenshot_dir, x)),
-            reverse=True
-        )
-        if screenshots:
-            return os.path.join(self.screenshot_dir, screenshots[0])
-        return None
-
-    def create_prompt_vars(self, current_action_description, next_goal):
-        prompt = {
-            "current_action_description": current_action_description,
-            "next_goal": next_goal,
-        }
-        return prompt
-
-    def send_screenshot_to_llm(self, screenshot_path, current_action_description="", next_goal=""):
-        with open(screenshot_path, "rb") as image_file:
-            encoded_image = base64.b64encode(image_file.read()).decode('utf-8')
-        
-        prompt_vars = self.create_prompt(current_action_description, next_goal)
-        response = self.send_prompt_to_llm(prompt_vars, encoded_image)  # Sending prompt and image separately
-        return json.loads(response)
-
-    def send_prompt_to_llm(self, prompt_vars, screenshot_base64):
-        raise NotImplementedError("This method should be implemented by subclasses: `def send_prompt_to_llm(self, prompt_vars, screenshot_base64)`")
-
-    def close(self):
-        self.save_cookies()
-        self.browser.close()
-        self.playwright.stop()
-
-    def execute_instructions(self, instructions):
-        for instruction in instructions:
-            action = instruction['action']
-            if action == 'navigate':
-                self.navigate(instruction['url'])
-            elif action == 'click':
-                self.click(instruction['selector'])
-            elif action == 'scroll':
-                self.scroll(instruction.get('direction', 'down'), instruction.get('amount', 1))
-            elif action == 'type':
-                self.type_text(instruction['selector'], instruction['text'])
-            screenshot_path = self.take_screenshot()
-            new_instructions = self.send_screenshot_to_llm(screenshot_path, instruction.get('description', ''), instruction.get('next_goal', ''))
-            if new_instructions:
-                self.execute_instructions(new_instructions)
-