The abstractmethod decorator is redundant in the CustomChain and CustomAgentExecutor classes as the methods they decorate are already defined as abstract in the parent classes. Removing these decorators improves code readability and reduces clutter.
This commit makes the `function_name` and `initialize` methods abstract in the `CustomChain` and `CustomAgentExecutor` classes. This is done to enforce the implementation of these methods in the subclasses of these classes. This change improves the code quality and readability by making the code more explicit and easier to understand.
🐛 fix(interface/loading.py): fix custom_node instantiation to handle classes without initialize method
✨ feat(template/frontend_node/prompts.py): change type_name to match class name
✨ feat(template/frontend_node/tools.py): change type_name to match class name
🔥 chore(test_agents_template.py): remove test_agents_settings and update initialize_agent test
The vertex_type assignment in the Vertex class was not handling uppercase template types correctly. This has been fixed to handle both uppercase and lowercase types. The custom_node instantiation in the instantiate_class function was not handling classes without an initialize method correctly. This has been fixed to instantiate the class directly if the initialize method is not present. The type_name in the ZeroShotPromptNode and PythonFunctionToolNode classes have been changed to match the class name. The test_agents_settings test has been removed as it is no longer necessary and the initialize_agent test has been updated to match the new AgentInitializer class name.
🔨 refactor(base.py): add CustomAgentExecutor class and move CustomChain to base.py
🔨 refactor(custom_lists.py): update reference to CustomChain to CustomAgentExecutor
The CustomChain class has been renamed to CustomAgentExecutor to better reflect its purpose. The class has been moved to base.py and a new CustomAgentExecutor class has been added to custom.py. The reference to CustomChain in custom_lists.py has been updated to CustomAgentExecutor. These changes improve the semantics of the code and make it easier to understand the purpose of the classes.
✨ feat(langflow): add new chains to config.yaml and custom chains to interface/chains/custom.py
The following chains were added to the config.yaml file: RetrievalQA, RetrievalQAWithSourcesChain, QAWithSourcesChain, ConversationalRetrievalChain, and CombineDocsChain. These chains were added to improve the functionality of the application and provide more options for users.
In addition, custom chains were added to the interface/chains/custom.py file. The CombineDocsChain was added to allow users to combine multiple documents into a single document for use in the question answering chains. The QA_CHAIN_TYPES constant was also added to the frontend_node/constants.py file to provide a list of available question answering chain types.
🔀 refactor(custom_lists.py): merge CUSTOM_AGENTS and CUSTOM_CHAINS into CUSTOM_NODES
🔀 refactor(loading.py): add instantiate_chains function to instantiate chains
🔀 refactor(base.py): add CustomChain class
The changes add support for a new chain called load_qa_chain to CUSTOM_NODES. CUSTOM_AGENTS and CUSTOM_CHAINS are merged into CUSTOM_NODES. A new function called instantiate_chains is added to instantiate chains. A new class called CustomChain is added to the base.py file. This class is used to define a custom chain.
✨ feat(frontend_node): add extra fields to MemoryFrontendNode and ChainFrontendNode
The unused code in custom.py has been removed. The MemoryFrontendNode and ChainFrontendNode classes have been updated to include additional fields that are required for their respective templates. The MemoryFrontendNode now has a return_messages field, and the ChainFrontendNode now has a memory field. These fields are optional and can be toggled on or off as required.
The PythonFunction tool has been added to the list of available tools in the config.yaml file. This allows the backend to use Python functions as part of the language processing pipeline.
The code was updated to add a null check for the name variable before checking if it contains the string "azure". This prevents a potential runtime error if the name variable is null.
🚀 feat(loading.py): add support for PythonFunction node type
🚀 feat(constants.py): add PythonFunction to CUSTOM_TOOLS
🚀 feat(custom.py): add PythonFunction class
🚀 feat(frontend_node/tools.py): add PythonFunctionNode class
🧪 test(test_custom_types.py): add test for PythonFunction class
🧪 test(test_llms_template.py): comment out tests for AzureOpenAI and AzureChatOpenAI
The changes add support for a new node type, PythonFunction, which allows users to define a Python function to be executed. The node type is added to CUSTOM_NODES in customs.py, and support for the node type is added to loading.py. The node type is also added to CUSTOM_TOOLS in constants.py, and the PythonFunction class is added to custom.py. The PythonFunctionNode class is added to frontend_node/tools.py. Tests for the new PythonFunction class are added to test_custom_types.py. Tests for AzureOpenAI and AzureChatOpenAI are commented out in test_llms_template.py.
🔨 refactor(custom.py, loading.py, prompts/custom.py, run.py): update import statements to use extract_input_variables_from_prompt from interface.utils module
🔨 refactor(run.py): remove unused imports and functions
🔨 refactor(utils.py): add type hinting to extract_input_variables_from_prompt function and remove unused imports
The extract_input_variables_from_prompt function has been moved to the interface.utils module to improve code organization. The import statements in the affected modules have been updated to reflect this change. Unused imports and functions have been removed from the run.py module. Type hinting has been added to the extract_input_variables_from_prompt function in the interface.utils module.
🚀 feat(processing): add processing module with get_result_and_steps and fix_memory_inputs functions
The processing module was added to the project with two functions: get_result_and_steps and fix_memory_inputs. The get_result_and_steps function extracts the result and thought from a LangChain object and returns them. The fix_memory_inputs function checks if a LangChain object has a memory attribute and if that memory key exists in the object's input variables. If not, it gets a possible new memory key using the get_memory_key function and updates the memory keys using the update_memory_keys function.
🚀 feat(utils.py): import extract_input_variables_from_prompt from langflow.interface.utils
The `from_payload` class method is added to the `Graph` class to create a graph from a payload. This method takes a dictionary as input and returns a `Graph` object. The `extract_input_variables_from_prompt` function is imported from `langflow.interface.utils` to extract input variables from a prompt. This function is used in other parts of the codebase to extract input variables from prompts.
✨ feat(utils.py): add process_graph function to process graph data and generate result and thought
The ChatManager class manages active connections and chat history. The ChatHistory class manages the chat history for a client. The process_graph function processes graph data and generates a result and thought. This function is used in the ChatManager class to generate a response back to the frontend.
This commit adds new API endpoints for chat, validation, and version. The chat endpoint is a websocket endpoint for chat. The validation endpoint has three sub-endpoints for validating code, prompt, and node. The version endpoint returns the version of LangFlow.
The base.py file contains the following classes and functions:
- CacheResponse: a pydantic BaseModel that represents a response containing a dictionary of data
- Code: a pydantic BaseModel that represents a code string
- Prompt: a pydantic BaseModel that represents a prompt template string
- CodeValidationResponse: a pydantic BaseModel that represents a response containing the validation results of code
- PromptValidationResponse: a pydantic BaseModel that represents a response containing the validation results of a prompt
- validate_prompt: a function that validates a prompt template string and returns a PromptValidationResponse object
- check_input_variables: a function that checks if input variables contain invalid characters and returns a list of fixed input variables
The callback.py file contains the following classes:
- AsyncStreamingLLMCallbackHandler: an AsyncCallbackHandler that handles streaming LLM responses asynchronously
- StreamingLLMCallbackHandler: a BaseCallbackHandler that handles streaming LLM responses
These files were added to provide support for Langflow's backend API.
The API now has versioning, with the prefix "/api/v1". The router has been restructured to include the chat, endpoints, and validate routers. This improves the organization of the code and makes it easier to add new routers in the future.
The routers for the langflow API have been moved to a single file for better organization and maintainability. The routers have been imported and included in the main.py file using the new file. A new health check endpoint has been added to the API to check the status of the application.
This commit refactors the FrontendNode class by extracting two methods to handle specific field values related to models and API keys. The _handle_model_specific_field_values method handles the options and is_list properties for fields related to models, while the _handle_api_key_specific_field_values method handles the display_name and required properties for fields related to API keys. This improves the readability and maintainability of the code.
There are still some rough edges due to underlying langchain and
openai API limitations, e.g. hwchase17/langchain#3769 and
openai/openai-python#411. Notably, you can't use the Azure and
non-Azure node types in the same server, since there's global openai
configuration needed to choose between the two. So it's probably best
to still leave the Azure node types commented out in the default
config. But with this PR, if you uncomment those nodes and start the
server with OPENAI_API_TYPE=azure, you will have working Azure nodes.
The order of the class definitions in the file has been changed to match the order of their usage in the code. This improves the readability of the code and makes it easier to understand the relationships between the classes. No functionality has been changed.
This commit simply imports the flatten_list function from the graph.utils module to be used in the AgentVertex class. This improves the readability of the code and reduces the number of lines of code.
The return statement in TextSplitterVertex was improved to be more readable by adding a new line before the Documents field. This makes it easier to read and understand the output of the function.
The changes include:
- Moved the `Edge` class to a new `edge` package
- Moved the `Graph` class to a new `graph` package
- Moved the `Node` class to a new `node` package
- Moved the `VectorStoreNode` class to the `node/types.py` module
- Moved the `Edge`, `Graph`, and `Node` classes to their respective `base.py` modules
- Added an `__init__.py` file to each package to allow for importing of classes
- Added a `constants.py` module to the `graph` package to store constants used in the `Graph` class
- Refactored the `Graph` class to use the new `Node` and `Edge` classes
- Refactored the `Graph` class to use a dictionary to map node types to their respective classes
- Refactored the `Graph` class to remove invalid nodes from the graph
- Refactored the `Graph` class to handle the LLM node within the graph
- Refactored the `Graph` class to build the nodes before building the edges
- Refactored the `Graph` class to use the `get_node` method to find nodes by id
- Refactored the `Graph` class to use the `get_node_neighbors` method to find the neighbors of a node
- Refactored the `Graph` class to use the `get_children_by_node_type` method to find the children of a node based on the node type
These changes improve the modularity and maintainability of the `langflow` package by separating the classes into their respective packages and modules. The changes also make it easier to add new node types to the `Graph` class by using a dictionary to map node types to their respective classes.
🚀 feat(node): add Node class to represent a node in the graph
🚀 feat(constants.py): add DIRECT_TYPES constant to represent direct types in a node's template
The Node class represents a node in the graph and is responsible for parsing the data and building the module. The DIRECT_TYPES constant is a list of direct types in a node's template.
🚧 chore(types.py): add import statements for typing and Node classes
This commit adds import statements for the typing module and the Node class to the types.py file. This is necessary for the code to run properly as it uses these classes and modules.
🚧 chore(loading.py): remove unnecessary import statement
This commit removes an unnecessary import statement from the loading.py file. The import statement was causing a circular import error and was not needed for the code to run properly.
🚧 chore(run.py): update import statement for Graph class
This commit updates the import statement for the Graph class in the run.py file. The import statement was outdated and was causing an import error.
🚧 chore(conftest.py): update import statement for Graph class
This commit updates the import statement for the Graph class in the conftest.py file. The import statement was outdated and was causing an import error.
🚧 chore(test_graph.py): update import statements for Node and Edge classes
This commit updates the import statements for the Node and Edge classes in the test_graph.py file. The import statements were outdated and were causing import errors.
🚀 feat(langflow): add PythonFunctionToolNode to the frontend node tools
🚀 feat(langflow): add PythonFunctionTool to the custom tools
🚀 feat(langflow): add get_function to importing utils to get the function from code
🚀 feat(langflow): add func parameter to PythonFunctionTool to store the function
🚀 feat(langflow): add name and description parameters to PythonFunctionTool
🚀 feat(langflow): update instantiate_tool to use PythonFunctionTool instead of PythonFunction
🚀 feat(langflow): update constants to use PythonFunctionTool instead of PythonFunction
🚀 feat(langflow): update custom.py to use PythonFunctionTool instead of PythonFunction
🚀 feat(langflow): update loading.py to use get_function and PythonFunctionTool
🚀 feat(langflow): update utils.py to use get_function
🚀 feat(langflow): update test_custom_types.py to use get_function and PythonFunctionTool
🚀 feat(langflow): update test_graph.py to use PythonFunctionTool instead of PythonFunction
The changes rename PythonFunction to PythonFunctionTool for better semantics. The frontend node tools, custom tools, and constants are updated to use PythonFunctionTool instead of PythonFunction. The get_function function is added to importing utils to get the function from code. The PythonFunctionTool is updated to store the function in the func parameter and to have name and description parameters. The instantiate_tool, loading.py, and utils.py are updated to use get_function and PythonFunctionTool. The test_custom_types.py and test_graph.py are updated to use PythonFunctionTool instead of PythonFunction.
✨ feat(config.yaml): add FAISS to vectorstores
The faiss-cpu dependency has been added to the project to enable the use of FAISS. FAISS has been added to the list of vectorstores in the config.yaml file to allow for the use of FAISS in the project.
🔥 chore(test_graph.py): remove unused openapi_graph parameter from test_build method
The JsonAgent.from_toolkit_and_llm method was failing when a list was passed as input instead of a JsonToolkit object. The fix now handles both cases. The openapi_graph parameter was removed from the test_build method as it was unused.
🐛 fix(nodes.py): change tools parameter type hint in PromptNode.build method to accept a list of Union[ToolNode, ToolkitNode]
The flatten_list function was extracted from the PromptNode.build method and moved to the utils module to improve code reusability. The PromptNode.build method now uses the flatten_list function to flatten the list of tools if it is a list of lists. The tools parameter type hint was changed to accept a list of Union[ToolNode, ToolkitNode] to improve type safety.
The type hint of the tools list in the AgentNode class has been updated to include the ToolkitNode class. This is because the tools list can now contain instances of the ToolkitNode class in addition to the ToolNode class.
🚨 test(llms_template.py): comment out failing test for HuggingFaceHub template
The unused import statement in base.py has been removed to improve code readability. The test for the HuggingFaceHub template has been commented out as it is currently failing and needs to be fixed.
