Module core.action_engine

Classes

class ActionEngine (driver: BaseDriver,
navigation_engine: BaseEngine = None,
python_engine: BaseEngine = None,
navigation_control: BaseEngine = None,
llm: BaseLLM = None,
embedding: BaseEmbedding = None,
retriever: BaseHtmlRetriever = None,
prompt_template: PromptTemplate = PromptTemplate(metadata={'prompt_type': <PromptType.CUSTOM: 'custom'>}, template_vars=['driver_capability', 'context_str', 'authorized_xpaths', 'query_str'], kwargs={}, output_parser=None, template_var_mappings=None, function_mappings=None, template='\n{driver_capability}\n\nHere is a the next example to answer:\n\nHTML:\n{context_str}\nAuthorized Xpaths: {authorized_xpaths}\nQuery: {query_str}\nCompletion:\n\n'),
extractor: BaseExtractor = <lavague.core.extractors.DynamicExtractor object>,
time_between_actions: float = 1.5,
n_attempts: int = 5,
logger: AgentLogger = None,
extraction_llm: Optional[BaseLLM] = None)

ActionEngine is a wrapper that instantiate every other engines (Navigation Engine, Python Engine, Navigation Control)

Args

driver (BaseDriver):

The Web driver used to interact with the headless browser

python_engine (BaseActionEngine)

Python Engine for generating code that doesn't interact with an html page.

navigation_control (BaseActionEngine)

Navigation Control

llm (BaseLLM)

llama-index LLM that will generate the action

embedding (BaseEmbedding)

llama-index Embedding model

retriever (BaseHtmlRetriever)

Specify which algorithm will be used for RAG

prompt_template (PromptTemplate)

Squelette of the final prompt

extractor (BaseExtractor)

Specify how to extract the final code from the llm answer

time_between_actions (float)

Time between each action

logger

(AgentLogger) Logger to log the actions taken by the agent

Static methods

def from_context(context: Context,
driver: BaseDriver,
navigation_engine: BaseEngine = None,
python_engine: BaseEngine = None,
navigation_control: BaseEngine = None,
retriever: BaseHtmlRetriever = None,
prompt_template: PromptTemplate = PromptTemplate(metadata={'prompt_type': <PromptType.CUSTOM: 'custom'>}, template_vars=['driver_capability', 'context_str', 'authorized_xpaths', 'query_str'], kwargs={}, output_parser=None, template_var_mappings=None, function_mappings=None, template='\n{driver_capability}\n\nHere is a the next example to answer:\n\nHTML:\n{context_str}\nAuthorized Xpaths: {authorized_xpaths}\nQuery: {query_str}\nCompletion:\n\n'),
extractor: BaseExtractor = <lavague.core.extractors.DynamicExtractor object>,
time_between_actions: float = 1.5,
n_attempts: int = 5,
logger: AgentLogger = None) ‑> ActionEngine

Create an ActionEngine from a context

Methods

def dispatch_instruction(self, next_engine_name: str, instruction: str) ‑> lavague.core.base_engine.ActionResult

Dispatch the instruction to the appropriate ActionEngine

Args

next_engine_name (str): The name of the engine to call instruction (str): The instruction to perform

Return

bool: True if the code was executed without error Any: The output of the code

def dispatch_instruction_gradio(self, next_engine_name: str, instruction: str)

Dispatch the instruction to the appropriate ActionEngine

Args

next_engine_name (str): The name of the engine to call instruction (str): The instruction to perform

Return

bool: True if the code was executed without error Any: The output of the code

def get_embedding_name(self)

def get_llm_name(self)

def set_display_all(self, display: bool)

def set_gradio_mode_all(self, gradio_mode: bool, objective, url_input, image_display, history)

def set_logger_all(self, logger: AgentLogger)