sunholo 0.74.8__tar.gz → 0.75.0__tar.gz

{sunholo-0.74.8 → sunholo-0.75.0}/PKG-INFO

@@ -1,9 +1,9 @@
 Metadata-Version: 2.1
 Name: sunholo
-Version: 0.74.8
+Version: 0.75.0
 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/v0.74.8.tar.gz
+Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.75.0.tar.gz
 Author: Holosun ApS
 Author-email: multivac@sunholo.com
 License: Apache License, Version 2.0

{sunholo-0.74.8 → sunholo-0.75.0}/setup.py

@@ -1,7 +1,7 @@
 from setuptools import setup, find_packages
 
 # Define your base version
-version = '0.74.8'
+version = '0.75.0'
 
 setup(
     name='sunholo',

sunholo-0.75.0/sunholo/auth/__init__.py

@@ -0,0 +1,2 @@
+from .run import get_header
+from .gcloud import get_local_gcloud_token

sunholo-0.75.0/sunholo/auth/gcloud.py

@@ -0,0 +1,14 @@
+import subprocess
+
+def get_local_gcloud_token():
+    # Use gcloud credentials locally
+
+    return (
+        subprocess.run(
+            ["gcloud", "auth", "print-identity-token"],
+            stdout=subprocess.PIPE,
+            check=True,
+        )
+        .stdout.strip()
+        .decode()
+    ) 

{sunholo-0.74.8 → sunholo-0.75.0}/sunholo/auth/run.py

@@ -7,6 +7,7 @@ from ..utils.gcp import is_running_on_cloudrun
 from ..utils.api_key import has_multivac_api_key, get_multivac_api_key
 from ..logging import log
 from ..agents.route import route_vac
+from .gcloud import get_local_gcloud_token
 
 def get_run_url(vector_name=None):
 
@@ -33,20 +34,11 @@ def get_id_token(url: str) -> str:
         import google.oauth2.id_token  # type: ignore
         auth_req = google.auth.transport.requests.Request()
         log.info(f'Got id_token for {url}')
+
         return google.oauth2.id_token.fetch_id_token(auth_req, url)
-    else:
-        # Use gcloud credentials locally
-        import subprocess
 
-        return (
-            subprocess.run(
-                ["gcloud", "auth", "print-identity-token"],
-                stdout=subprocess.PIPE,
-                check=True,
-            )
-            .stdout.strip()
-            .decode()
-        )
+    return get_local_gcloud_token()
+
 
 def get_header(vector_name) -> Optional[dict]:
     if has_multivac_api_key():

{sunholo-0.74.8 → sunholo-0.75.0}/sunholo/tools/web_browser.py

@@ -213,6 +213,35 @@ class BrowseWebWithImagePromptsBot:
         except Exception as err:
             log.warning(f"navigate failed with {str(err)}")
             self.action_log.append(f"Tried to navigate to {url} but got an error")
+
+    def get_locator_via_roles_and_placeholder(self, selector: str):
+        interactive_roles = ["button", "link", "menuitem", "menuitemcheckbox", "menuitemradio", "tab", "option"]
+
+        for role in interactive_roles:
+            log.info(f'Trying role {role} for selector {selector}')
+            elements = self.page.get_by_role(role).get_by_placeholder(selector).locator("visible=true").all()
+            if elements:
+                log.info(f"Got {len(elements)} elements for selector {selector} with role {role}")
+                for element in elements:
+                    try:
+                        log.info(f"Trying {selector} with element.hover locator: {element}")
+                        try:
+                            element.hover(timeout=10000, trial=True)
+                            self.action_log.append(f"Successfully found element via selector: {selector}")
+
+                            return element
+                        
+                        except Exception as err:
+                            log.warning(f"Could not hover over element: {element} {str(err)} - trying next element")
+                    except Exception as e:
+                        log.error(f"Failed to get locator for selector '{selector}' with role {role}: {str(e)}")
+                    
+                    time.sleep(0.5)  # Wait for a bit before retrying
+            
+            log.info(f"No elements for '{selector}' within role '{role}'")
+
+        self.action_log.append(f"FAILED: Using page.get_by_role('role').get_by_placeholder('{selector}').locator('visible=true') could not find any valid element. Try something else.")
+        return None
     
     def get_locator_via_roles_and_text(self, selector: str):
         interactive_roles = ["button", "link", "menuitem", "menuitemcheckbox", "menuitemradio", "tab", "option"]
@@ -240,7 +269,7 @@ class BrowseWebWithImagePromptsBot:
             
             log.info(f"No elements for '{selector}' within role '{role}'")
 
-        self.action_log.append(f"FAILED: Using page.get_by_role('role').locator('text={selector}').locator('visible=true') could not find any valid element. Try something else.")
+        self.action_log.append(f"FAILED: Using page.get_by_role('role').get_by_text('{selector}').locator('visible=true') could not find any valid element. Try something else.")
         return None
 
     def get_locator(self, selector, by_text=True):
@@ -256,10 +285,10 @@ class BrowseWebWithImagePromptsBot:
 
         return None
 
-    def click(self, selector, by_text=True):
+    def click(self, selector):
         (x,y)=(0,0)
 
-        element = self.get_locator(selector, by_text=by_text)
+        element = self.get_locator_via_roles_and_text(selector)
         if element is None:
             self.action_log.append(f"Tried to click on text {selector} but it was not a valid location to click")
             return (x,y)
@@ -303,9 +332,9 @@ class BrowseWebWithImagePromptsBot:
             log.warning(f"Scrolled failed with {str(err)}")
             self.action_log.append(f"Tried to scroll {direction} by {amount} pixels but got an error")
 
-    def type_text(self, selector, text, by_text=True):
+    def type_text(self, selector, text):
         (x,y)=(0,0)
-        element = self.get_locator(selector, by_text=by_text)
+        element = self.get_locator_via_roles_and_placeholder(selector)
         if element is None:
             self.action_log.append(f"Tried to type {text} via website text: {selector} but it was not a valid element to add text")
             return (x,y)
@@ -332,6 +361,68 @@ class BrowseWebWithImagePromptsBot:
 
             return (x, y)
 
+    def execute_custom_command(self, command):
+        """
+        Executes a custom command on the page object.
+
+        Args:
+            command (str): The command string to be executed.
+        """
+        try:
+            element_part = command.get('get_locator')
+            operation = command.get('operation')
+
+            if not element_part or not operation:
+                raise ValueError("Both 'element_part' and 'operation' must be provided in the command")
+            
+            # Dynamically get the method and its parameters
+            method_name, params = self.parse_element_part(element_part)
+            method = getattr(self.page, method_name)
+            element = method(*params)
+
+            if not element:
+                raise ValueError(f"Element not found for selector: {element_part}")
+
+            # Execute the operation
+            exec(f"element.{operation}")
+
+            # Mark the action on the screenshot
+            bounding_box = element.bounding_box()
+            if bounding_box:
+                x = bounding_box['x'] + bounding_box['width'] / 2
+                y = bounding_box['y'] + bounding_box['height'] / 2
+                mark_action = {'type': operation, 'position': (x, y)}
+                self.take_screenshot(mark_action=mark_action)
+            else:
+                self.take_screenshot()
+
+            log.info(f"Executed custom command on element: {element_part} with operation: {operation}")
+            self.action_log.append(f"Executed custom command on element: {element_part} with operation: {operation}")
+
+        except Exception as e:
+            log.error(f"Failed to execute custom command: {command}. Error: {str(e)}")
+            self.action_log.append(f"Failed to execute custom command: {command}. Error: {str(e)}")
+
+    def parse_element_part(self, element_part):
+        """
+        Parses the element_part string to extract the method name and its parameters.
+
+        Args:
+            element_part (str): The element part string (e.g., "get_by_role('button')")
+
+        Returns:
+            tuple: A tuple containing the method name and a list of parameters.
+        """
+        try:
+            # Extract the method name and parameters
+            method_name = element_part.split('(')[0]
+            params_str = element_part.split('(')[1].rstrip(')')
+            params = eval(f'[{params_str}]')  # Safely evaluate parameters
+
+            return method_name, params
+        except Exception as e:
+            raise ValueError(f"Failed to parse element part: {element_part}. Error: {str(e)}")
+
     def take_screenshot(self, full_page=False, mark_action=None):
 
         from PIL import Image
@@ -344,6 +435,10 @@ class BrowseWebWithImagePromptsBot:
             url_path = "index.html"
         else:
             url_path = url_path.replace("/","_")
+        
+        if get_clean_website_name(url_path) != self.website_name:
+            url_path = f"{get_clean_website_name(url_path)}_{url_path}"
+            
         screenshot_path = os.path.join(self.screenshot_dir, f"{timestamp}_{url_path}.png")
         screenshot_bytes = self.page.screenshot(full_page=full_page, scale='css')
         
@@ -358,7 +453,7 @@ class BrowseWebWithImagePromptsBot:
         #self.action_log.append(f"Screenshot {self.page.url} taken and saved to {screenshot_path}")
         self.session_screenshots.append(screenshot_path)
 
-        return screenshot_path
+        return screenshot_bytes
 
     def mark_screenshot(self, screenshot_bytes, mark_action):
         """
@@ -433,9 +528,9 @@ class BrowseWebWithImagePromptsBot:
 
         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')
+    def send_screenshot_to_llm(self, screenshot_bytes, last_message):
+
+        encoded_image = base64.b64encode(screenshot_bytes).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
@@ -490,14 +585,18 @@ This method should be implemented by subclasses: `def send_prompt_to_llm(self, p
                 x,y = self.type_text(instruction['selector'], instruction['text'])
                 if (x,y) != (0,0):
                     mark_action = {'type':'type', 'position': (x,y)}
+            elif action == 'execute':
+                x,y,mark = self.execute_custom_command(instruction['command'])
+                if mark:
+                    mark_action = {'type': mark, 'position': (x,y)}
             self.steps += 1
             if self.steps >= self.max_steps:
                 log.warning(f"Reached the maximum number of steps: {self.max_steps}")
                 return
         time.sleep(2) 
-        screenshot_path = self.take_screenshot(mark_action=mark_action)
+        screenshot_bytes = self.take_screenshot(mark_action=mark_action)
         next_browser_instructions = self.send_screenshot_to_llm(
-                screenshot_path, 
+                screenshot_bytes, 
                 last_message=last_message)
             
         return next_browser_instructions

{sunholo-0.74.8 → sunholo-0.75.0}/sunholo/utils/parsers.py

@@ -186,6 +186,6 @@ def escape_braces(text):
     text = re.sub(r'(?<!})}(?!})', '}}', text)  # Replace '}' with '}}' if not already double braced
     return text
 
-def get_clean_website_name(url):
+def get_clean_website_name(url: str):
     parsed_url = urllib.parse.urlparse(url)
     return parsed_url.netloc

{sunholo-0.74.8 → sunholo-0.75.0}/sunholo/vertex/extensions_class.py

@@ -236,6 +236,7 @@ class VertexAIExtensions:
             if extension_name is None:
                 raise ValueError("Must specify extension_id or init one with class")
         else:  
+            extension_id = str(extension_id)
             if not extension_id.startswith("projects/"):
                 project_id = get_gcp_project()
                 extension_name = f"projects/{project_id}/locations/{self.location}/extensions/{extension_id}"
@@ -244,11 +245,29 @@ class VertexAIExtensions:
 
         extension = extensions.Extension(extension_name)
 
+        log.info(f"Executing extension {extension_name=} with {operation_id=} and {operation_params=}")
+
+        # local testing auth
+        from ..utils.gcp import is_running_on_cloudrun
+        auth_config=None # on cloud run it sorts itself out via default creds(?)
+
+        if not is_running_on_cloudrun():
+            from ..auth import get_local_gcloud_token
+            log.warning("Using local authentication via gcloud")
+            auth_config = {
+                    "authType": "OAUTH",
+                    "oauth_config": {"access_token": f"'{get_local_gcloud_token()}'"}
+                }
+            log.info(auth_config)
+
         response = extension.execute(
             operation_id=operation_id,
             operation_params=operation_params,
+            runtime_auth_config=auth_config
         )
 
+        log.info(f"Extension {extension_name=} {response=}")
+
         return response
 
     def execute_code_extension(self, 

{sunholo-0.74.8 → sunholo-0.75.0}/sunholo.egg-info/PKG-INFO

@@ -1,9 +1,9 @@
 Metadata-Version: 2.1
 Name: sunholo
-Version: 0.74.8
+Version: 0.75.0
 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/v0.74.8.tar.gz
+Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.75.0.tar.gz
 Author: Holosun ApS
 Author-email: multivac@sunholo.com
 License: Apache License, Version 2.0

{sunholo-0.74.8 → sunholo-0.75.0}/sunholo.egg-info/SOURCES.txt

@@ -29,6 +29,7 @@ sunholo/agents/flask/vac_routes.py
 sunholo/archive/__init__.py
 sunholo/archive/archive.py
 sunholo/auth/__init__.py
+sunholo/auth/gcloud.py
 sunholo/auth/run.py
 sunholo/bots/__init__.py
 sunholo/bots/discord.py

sunholo-0.74.8/sunholo/auth/__init__.py

@@ -1 +0,0 @@
-from .run import get_header