langgraph 0.1.10

Summary of Changes

• Added InjectedState for automatically injecting graph state into tool arguments in ToolNode #1067
• Improved parallel execution of tools in ToolNode using config lists #1066
• Refactored the internal Pregel graph processing engine for better performance and maintainability
• Enhanced checkpoint management with improved parent-child relationship handling
• Added new error classes for better handling of graph interruptions and empty inputs
• Improved comparison logic for BinaryOperatorAggregate to handle lambda operators
• Significantly enhanced documentation in checkpoint-related classes

Detailed Changes

langgraph.prebuilt.tool_node.InjectedState

• Added new InjectedState class that allows automatically injecting graph state into tool arguments #1067
• Enables tools to access specific parts of graph state without requiring models to generate them
• Similar to InjectedToolArg but specifically for accessing graph state

langgraph.prebuilt.tool_node.ToolNode

• Refactored ToolNode implementation to improve code organization and maintainability #1066
• Added support for parallel tool execution using get_config_list from langchain-core
• Split functionality into smaller methods (_run_one, _arun_one, _parse_input, etc.)
• Added state injection capability for tools that use the InjectedState annotation #1067

langgraph.channels.binop.BinaryOperatorAggregate

• Improved equality comparison to better handle lambda operators
• Fixed comparison logic to avoid false negatives when comparing instances with lambda functions

langgraph.checkpoint

• Enhanced documentation across checkpoint implementations (Memory, SQLite, AsyncSQLite)
• Improved typing and documentation for CheckpointMetadata and related classes
• Added better parent-child relationship tracking in checkpoints for nested graphs
• Enhanced checkpoint tuple to include parent configuration data

langgraph.errors

• Added GraphInterrupt class for handling interruptions in nested graphs
• Added EmptyInputError for better error reporting when graphs receive empty inputs

langgraph.pregel

• Significantly refactored the Pregel graph processing engine for better maintainability
• Split functionality into focused modules like algo.py and loop.py
• Improved handling of interruptions in nested graphs
• Enhanced the BackgroundExecutor and AsyncBackgroundExecutor classes for better resource management
• Added better checkpointing behavior, especially for nested graphs