optexity-browser-use 0.9.5__py3-none-any.whl

browser_use/llm/messages.py

@@ -0,0 +1,238 @@
+"""
+This implementation is based on the OpenAI types, while removing all the parts that are not needed for Browser Use.
+"""
+
+# region - Content parts
+from typing import Literal, Union
+
+from openai import BaseModel
+
+
+def _truncate(text: str, max_length: int = 50) -> str:
+	"""Truncate text to max_length characters, adding ellipsis if truncated."""
+	if len(text) <= max_length:
+		return text
+	return text[: max_length - 3] + '...'
+
+
+def _format_image_url(url: str, max_length: int = 50) -> str:
+	"""Format image URL for display, truncating if necessary."""
+	if url.startswith('data:'):
+		# Base64 image
+		media_type = url.split(';')[0].split(':')[1] if ';' in url else 'image'
+		return f'<base64 {media_type}>'
+	else:
+		# Regular URL
+		return _truncate(url, max_length)
+
+
+class ContentPartTextParam(BaseModel):
+	text: str
+	type: Literal['text'] = 'text'
+
+	def __str__(self) -> str:
+		return f'Text: {_truncate(self.text)}'
+
+	def __repr__(self) -> str:
+		return f'ContentPartTextParam(text={_truncate(self.text)})'
+
+
+class ContentPartRefusalParam(BaseModel):
+	refusal: str
+	type: Literal['refusal'] = 'refusal'
+
+	def __str__(self) -> str:
+		return f'Refusal: {_truncate(self.refusal)}'
+
+	def __repr__(self) -> str:
+		return f'ContentPartRefusalParam(refusal={_truncate(repr(self.refusal), 50)})'
+
+
+SupportedImageMediaType = Literal['image/jpeg', 'image/png', 'image/gif', 'image/webp']
+
+
+class ImageURL(BaseModel):
+	url: str
+	"""Either a URL of the image or the base64 encoded image data."""
+	detail: Literal['auto', 'low', 'high'] = 'auto'
+	"""Specifies the detail level of the image.
+
+    Learn more in the
+    [Vision guide](https://platform.openai.com/docs/guides/vision#low-or-high-fidelity-image-understanding).
+    """
+	# needed for Anthropic
+	media_type: SupportedImageMediaType = 'image/jpeg'
+
+	def __str__(self) -> str:
+		url_display = _format_image_url(self.url)
+		return f'🖼️  Image[{self.media_type}, detail={self.detail}]: {url_display}'
+
+	def __repr__(self) -> str:
+		url_repr = _format_image_url(self.url, 30)
+		return f'ImageURL(url={repr(url_repr)}, detail={repr(self.detail)}, media_type={repr(self.media_type)})'
+
+
+class ContentPartImageParam(BaseModel):
+	image_url: ImageURL
+	type: Literal['image_url'] = 'image_url'
+
+	def __str__(self) -> str:
+		return str(self.image_url)
+
+	def __repr__(self) -> str:
+		return f'ContentPartImageParam(image_url={repr(self.image_url)})'
+
+
+class Function(BaseModel):
+	arguments: str
+	"""
+    The arguments to call the function with, as generated by the model in JSON
+    format. Note that the model does not always generate valid JSON, and may
+    hallucinate parameters not defined by your function schema. Validate the
+    arguments in your code before calling your function.
+    """
+	name: str
+	"""The name of the function to call."""
+
+	def __str__(self) -> str:
+		args_preview = _truncate(self.arguments, 80)
+		return f'{self.name}({args_preview})'
+
+	def __repr__(self) -> str:
+		args_repr = _truncate(repr(self.arguments), 50)
+		return f'Function(name={repr(self.name)}, arguments={args_repr})'
+
+
+class ToolCall(BaseModel):
+	id: str
+	"""The ID of the tool call."""
+	function: Function
+	"""The function that the model called."""
+	type: Literal['function'] = 'function'
+	"""The type of the tool. Currently, only `function` is supported."""
+
+	def __str__(self) -> str:
+		return f'ToolCall[{self.id}]: {self.function}'
+
+	def __repr__(self) -> str:
+		return f'ToolCall(id={repr(self.id)}, function={repr(self.function)})'
+
+
+# endregion
+
+
+# region - Message types
+class _MessageBase(BaseModel):
+	"""Base class for all message types"""
+
+	role: Literal['user', 'system', 'assistant']
+
+	cache: bool = False
+	"""Whether to cache this message. This is only applicable when using Anthropic models.
+	"""
+
+
+class UserMessage(_MessageBase):
+	role: Literal['user'] = 'user'
+	"""The role of the messages author, in this case `user`."""
+
+	content: str | list[ContentPartTextParam | ContentPartImageParam]
+	"""The contents of the user message."""
+
+	name: str | None = None
+	"""An optional name for the participant.
+
+    Provides the model information to differentiate between participants of the same
+    role.
+    """
+
+	@property
+	def text(self) -> str:
+		"""
+		Automatically parse the text inside content, whether it's a string or a list of content parts.
+		"""
+		if isinstance(self.content, str):
+			return self.content
+		elif isinstance(self.content, list):
+			return '\n'.join([part.text for part in self.content if part.type == 'text'])
+		else:
+			return ''
+
+	def __str__(self) -> str:
+		return f'UserMessage(content={self.text})'
+
+	def __repr__(self) -> str:
+		return f'UserMessage(content={repr(self.text)})'
+
+
+class SystemMessage(_MessageBase):
+	role: Literal['system'] = 'system'
+	"""The role of the messages author, in this case `system`."""
+
+	content: str | list[ContentPartTextParam]
+	"""The contents of the system message."""
+
+	name: str | None = None
+
+	@property
+	def text(self) -> str:
+		"""
+		Automatically parse the text inside content, whether it's a string or a list of content parts.
+		"""
+		if isinstance(self.content, str):
+			return self.content
+		elif isinstance(self.content, list):
+			return '\n'.join([part.text for part in self.content if part.type == 'text'])
+		else:
+			return ''
+
+	def __str__(self) -> str:
+		return f'SystemMessage(content={self.text})'
+
+	def __repr__(self) -> str:
+		return f'SystemMessage(content={repr(self.text)})'
+
+
+class AssistantMessage(_MessageBase):
+	role: Literal['assistant'] = 'assistant'
+	"""The role of the messages author, in this case `assistant`."""
+
+	content: str | list[ContentPartTextParam | ContentPartRefusalParam] | None
+	"""The contents of the assistant message."""
+
+	name: str | None = None
+
+	refusal: str | None = None
+	"""The refusal message by the assistant."""
+
+	tool_calls: list[ToolCall] = []
+	"""The tool calls generated by the model, such as function calls."""
+
+	@property
+	def text(self) -> str:
+		"""
+		Automatically parse the text inside content, whether it's a string or a list of content parts.
+		"""
+		if isinstance(self.content, str):
+			return self.content
+		elif isinstance(self.content, list):
+			text = ''
+			for part in self.content:
+				if part.type == 'text':
+					text += part.text
+				elif part.type == 'refusal':
+					text += f'[Refusal] {part.refusal}'
+			return text
+		else:
+			return ''
+
+	def __str__(self) -> str:
+		return f'AssistantMessage(content={self.text})'
+
+	def __repr__(self) -> str:
+		return f'AssistantMessage(content={repr(self.text)})'
+
+
+BaseMessage = Union[UserMessage, SystemMessage, AssistantMessage]
+
+# endregion

browser_use/llm/models.py

@@ -0,0 +1,271 @@
+"""
+Convenient access to LLM models.
+
+Usage:
+    from browser_use import llm
+
+    # Simple model access
+    model = llm.azure_gpt_4_1_mini
+    model = llm.openai_gpt_4o
+    model = llm.google_gemini_2_5_pro
+    model = llm.bu_latest
+"""
+
+import os
+from typing import TYPE_CHECKING
+
+from browser_use.llm.azure.chat import ChatAzureOpenAI
+from browser_use.llm.browser_use.chat import ChatBrowserUse
+from browser_use.llm.cerebras.chat import ChatCerebras
+from browser_use.llm.google.chat import ChatGoogle
+from browser_use.llm.openai.chat import ChatOpenAI
+
+# Optional OCI import
+try:
+	from browser_use.llm.oci_raw.chat import ChatOCIRaw
+
+	OCI_AVAILABLE = True
+except ImportError:
+	ChatOCIRaw = None
+	OCI_AVAILABLE = False
+
+if TYPE_CHECKING:
+	from browser_use.llm.base import BaseChatModel
+
+# Type stubs for IDE autocomplete
+openai_gpt_4o: 'BaseChatModel'
+openai_gpt_4o_mini: 'BaseChatModel'
+openai_gpt_4_1_mini: 'BaseChatModel'
+openai_o1: 'BaseChatModel'
+openai_o1_mini: 'BaseChatModel'
+openai_o1_pro: 'BaseChatModel'
+openai_o3: 'BaseChatModel'
+openai_o3_mini: 'BaseChatModel'
+openai_o3_pro: 'BaseChatModel'
+openai_o4_mini: 'BaseChatModel'
+openai_gpt_5: 'BaseChatModel'
+openai_gpt_5_mini: 'BaseChatModel'
+openai_gpt_5_nano: 'BaseChatModel'
+
+azure_gpt_4o: 'BaseChatModel'
+azure_gpt_4o_mini: 'BaseChatModel'
+azure_gpt_4_1_mini: 'BaseChatModel'
+azure_o1: 'BaseChatModel'
+azure_o1_mini: 'BaseChatModel'
+azure_o1_pro: 'BaseChatModel'
+azure_o3: 'BaseChatModel'
+azure_o3_mini: 'BaseChatModel'
+azure_o3_pro: 'BaseChatModel'
+azure_gpt_5: 'BaseChatModel'
+azure_gpt_5_mini: 'BaseChatModel'
+
+google_gemini_2_0_flash: 'BaseChatModel'
+google_gemini_2_0_pro: 'BaseChatModel'
+google_gemini_2_5_pro: 'BaseChatModel'
+google_gemini_2_5_flash: 'BaseChatModel'
+google_gemini_2_5_flash_lite: 'BaseChatModel'
+
+cerebras_llama3_1_8b: 'BaseChatModel'
+cerebras_llama3_3_70b: 'BaseChatModel'
+cerebras_gpt_oss_120b: 'BaseChatModel'
+cerebras_llama_4_scout_17b_16e_instruct: 'BaseChatModel'
+cerebras_llama_4_maverick_17b_128e_instruct: 'BaseChatModel'
+cerebras_qwen_3_32b: 'BaseChatModel'
+cerebras_qwen_3_235b_a22b_instruct_2507: 'BaseChatModel'
+cerebras_qwen_3_235b_a22b_thinking_2507: 'BaseChatModel'
+cerebras_qwen_3_coder_480b: 'BaseChatModel'
+
+bu_latest: 'BaseChatModel'
+bu_1_0: 'BaseChatModel'
+
+
+def get_llm_by_name(model_name: str):
+	"""
+	Factory function to create LLM instances from string names with API keys from environment.
+
+	Args:
+	    model_name: String name like 'azure_gpt_4_1_mini', 'openai_gpt_4o', etc.
+
+	Returns:
+	    LLM instance with API keys from environment variables
+
+	Raises:
+	    ValueError: If model_name is not recognized
+	"""
+	if not model_name:
+		raise ValueError('Model name cannot be empty')
+
+	# Parse model name
+	parts = model_name.split('_', 1)
+	if len(parts) < 2:
+		raise ValueError(f"Invalid model name format: '{model_name}'. Expected format: 'provider_model_name'")
+
+	provider = parts[0]
+	model_part = parts[1]
+
+	# Convert underscores back to dots/dashes for actual model names
+	if 'gpt_4_1_mini' in model_part:
+		model = model_part.replace('gpt_4_1_mini', 'gpt-4.1-mini')
+	elif 'gpt_4o_mini' in model_part:
+		model = model_part.replace('gpt_4o_mini', 'gpt-4o-mini')
+	elif 'gpt_4o' in model_part:
+		model = model_part.replace('gpt_4o', 'gpt-4o')
+	elif 'gemini_2_0' in model_part:
+		model = model_part.replace('gemini_2_0', 'gemini-2.0').replace('_', '-')
+	elif 'gemini_2_5' in model_part:
+		model = model_part.replace('gemini_2_5', 'gemini-2.5').replace('_', '-')
+	elif 'llama3_1' in model_part:
+		model = model_part.replace('llama3_1', 'llama3.1').replace('_', '-')
+	elif 'llama3_3' in model_part:
+		model = model_part.replace('llama3_3', 'llama-3.3').replace('_', '-')
+	elif 'llama_4_scout' in model_part:
+		model = model_part.replace('llama_4_scout', 'llama-4-scout').replace('_', '-')
+	elif 'llama_4_maverick' in model_part:
+		model = model_part.replace('llama_4_maverick', 'llama-4-maverick').replace('_', '-')
+	elif 'gpt_oss_120b' in model_part:
+		model = model_part.replace('gpt_oss_120b', 'gpt-oss-120b')
+	elif 'qwen_3_32b' in model_part:
+		model = model_part.replace('qwen_3_32b', 'qwen-3-32b')
+	elif 'qwen_3_235b_a22b_instruct' in model_part:
+		if model_part.endswith('_2507'):
+			model = model_part.replace('qwen_3_235b_a22b_instruct_2507', 'qwen-3-235b-a22b-instruct-2507')
+		else:
+			model = model_part.replace('qwen_3_235b_a22b_instruct', 'qwen-3-235b-a22b-instruct-2507')
+	elif 'qwen_3_235b_a22b_thinking' in model_part:
+		if model_part.endswith('_2507'):
+			model = model_part.replace('qwen_3_235b_a22b_thinking_2507', 'qwen-3-235b-a22b-thinking-2507')
+		else:
+			model = model_part.replace('qwen_3_235b_a22b_thinking', 'qwen-3-235b-a22b-thinking-2507')
+	elif 'qwen_3_coder_480b' in model_part:
+		model = model_part.replace('qwen_3_coder_480b', 'qwen-3-coder-480b')
+	else:
+		model = model_part.replace('_', '-')
+
+	# OpenAI Models
+	if provider == 'openai':
+		api_key = os.getenv('OPENAI_API_KEY')
+		return ChatOpenAI(model=model, api_key=api_key)
+
+	# Azure OpenAI Models
+	elif provider == 'azure':
+		api_key = os.getenv('AZURE_OPENAI_KEY') or os.getenv('AZURE_OPENAI_API_KEY')
+		azure_endpoint = os.getenv('AZURE_OPENAI_ENDPOINT')
+		return ChatAzureOpenAI(model=model, api_key=api_key, azure_endpoint=azure_endpoint)
+
+	# Google Models
+	elif provider == 'google':
+		api_key = os.getenv('GOOGLE_API_KEY')
+		return ChatGoogle(model=model, api_key=api_key)
+
+	# OCI Models
+	elif provider == 'oci':
+		# OCI requires more complex configuration that can't be easily inferred from env vars
+		# Users should use ChatOCIRaw directly with proper configuration
+		raise ValueError('OCI models require manual configuration. Use ChatOCIRaw directly with your OCI credentials.')
+
+	# Cerebras Models
+	elif provider == 'cerebras':
+		api_key = os.getenv('CEREBRAS_API_KEY')
+		return ChatCerebras(model=model, api_key=api_key)
+
+	# Browser Use Models
+	elif provider == 'bu':
+		# Handle bu_latest -> bu-latest conversion (need to prepend 'bu-' back)
+		model = f'bu-{model_part.replace("_", "-")}'
+		api_key = os.getenv('BROWSER_USE_API_KEY')
+		return ChatBrowserUse(model=model, api_key=api_key)
+
+	else:
+		available_providers = ['openai', 'azure', 'google', 'oci', 'cerebras', 'bu']
+		raise ValueError(f"Unknown provider: '{provider}'. Available providers: {', '.join(available_providers)}")
+
+
+# Pre-configured model instances (lazy loaded via __getattr__)
+def __getattr__(name: str) -> 'BaseChatModel':
+	"""Create model instances on demand with API keys from environment."""
+	# Handle chat classes first
+	if name == 'ChatOpenAI':
+		return ChatOpenAI  # type: ignore
+	elif name == 'ChatAzureOpenAI':
+		return ChatAzureOpenAI  # type: ignore
+	elif name == 'ChatGoogle':
+		return ChatGoogle  # type: ignore
+	elif name == 'ChatOCIRaw':
+		if not OCI_AVAILABLE:
+			raise ImportError('OCI integration not available. Install with: pip install "browser-use[oci]"')
+		return ChatOCIRaw  # type: ignore
+	elif name == 'ChatCerebras':
+		return ChatCerebras  # type: ignore
+	elif name == 'ChatBrowserUse':
+		return ChatBrowserUse  # type: ignore
+
+	# Handle model instances - these are the main use case
+	try:
+		return get_llm_by_name(name)
+	except ValueError:
+		raise AttributeError(f"module '{__name__}' has no attribute '{name}'")
+
+
+# Export all classes and preconfigured instances, conditionally including ChatOCIRaw
+__all__ = [
+	'ChatOpenAI',
+	'ChatAzureOpenAI',
+	'ChatGoogle',
+	'ChatCerebras',
+	'ChatBrowserUse',
+]
+
+if OCI_AVAILABLE:
+	__all__.append('ChatOCIRaw')
+
+__all__ += [
+	'get_llm_by_name',
+	# OpenAI instances - created on demand
+	'openai_gpt_4o',
+	'openai_gpt_4o_mini',
+	'openai_gpt_4_1_mini',
+	'openai_o1',
+	'openai_o1_mini',
+	'openai_o1_pro',
+	'openai_o3',
+	'openai_o3_mini',
+	'openai_o3_pro',
+	'openai_o4_mini',
+	'openai_gpt_5',
+	'openai_gpt_5_mini',
+	'openai_gpt_5_nano',
+	# Azure instances - created on demand
+	'azure_gpt_4o',
+	'azure_gpt_4o_mini',
+	'azure_gpt_4_1_mini',
+	'azure_o1',
+	'azure_o1_mini',
+	'azure_o1_pro',
+	'azure_o3',
+	'azure_o3_mini',
+	'azure_o3_pro',
+	'azure_gpt_5',
+	'azure_gpt_5_mini',
+	# Google instances - created on demand
+	'google_gemini_2_0_flash',
+	'google_gemini_2_0_pro',
+	'google_gemini_2_5_pro',
+	'google_gemini_2_5_flash',
+	'google_gemini_2_5_flash_lite',
+	# Cerebras instances - created on demand
+	'cerebras_llama3_1_8b',
+	'cerebras_llama3_3_70b',
+	'cerebras_gpt_oss_120b',
+	'cerebras_llama_4_scout_17b_16e_instruct',
+	'cerebras_llama_4_maverick_17b_128e_instruct',
+	'cerebras_qwen_3_32b',
+	'cerebras_qwen_3_235b_a22b_instruct_2507',
+	'cerebras_qwen_3_235b_a22b_thinking_2507',
+	'cerebras_qwen_3_coder_480b',
+	# Browser Use instances - created on demand
+	'bu_latest',
+	'bu_1_0',
+]
+
+# NOTE: OCI backend is optional. The try/except ImportError and conditional __all__ are required
+# so this module can be imported without browser-use[oci] installed.

browser_use/llm/oci_raw/__init__.py

@@ -0,0 +1,10 @@
+"""
+OCI Raw API integration for browser-use.
+
+This module provides direct integration with Oracle Cloud Infrastructure's
+Generative AI service using the raw API endpoints, without Langchain dependencies.
+"""
+
+from .chat import ChatOCIRaw
+
+__all__ = ['ChatOCIRaw']