0.2.44

langgraph 0.2.44

Summary of Changes

• Added validation to ensure all tool calls in AI messages have corresponding Tool messages in chat history, improving error handling for ReAct agents #2182
• Improved graph visualization by better handling node names when they're not explicitly specified #2304
• Enhanced stream mode handling for remote graphs to support messages-tuple format #2304

Detailed Changes

langgraph.errors.ErrorCode

• Added a new error code INVALID_CHAT_HISTORY to help identify and handle issues with incomplete chat histories #2182

langgraph.prebuilt.chat_agent_executor

• Added new validation function _validate_chat_history that checks if all tool calls in AIMessages have corresponding ToolMessages #2182
• Integrated this validation in both synchronous and asynchronous model calling functions to ensure proper chat history structure before sending to LLM #2182
• Provides clear error messages when validation fails, showing examples of problematic tool calls #2182

langgraph.pregel.remote.RemoteGraph

• Improved node name resolution logic in _get_drawable_nodes to provide more meaningful names in visualizations #2304
• Now tries to find names from node data if not explicitly specified, and falls back to node ID instead of empty string #2304
• Added support for "messages-tuple" format in stream modes by mapping it to "messages" in requested stream modes #2304