Support for starting Graph execution from any node with pre-existing State

[siddarthreddygsr]

Currently, LangGraph’s execution always starts from a designated entry point (e.g., the START node), and while it supports passing a pre-existing state, there is no direct way to begin execution at an arbitrary node using that state. This can limit flexibility in certain use cases—such as debugging, checkpointing, or resuming workflows from specific points.

Add an API method (e.g., start_from_node(node_id, state)) that allows users to specify both the starting node and the initial state. This would enable more advanced use cases, including:

• Debugging workflows by starting execution at any node.
• Resuming workflows from a checkpointed state at a specific node for chat applications where chat flow is defined using langgraph.

[hinthornw]

Does this work? (just look at field )

def route_entry(state):
        return state["start_here"]
        
graph = StateGraph(...).add_conditional_edges("__start__", route_entry)... etc.

I'm not yet convinced another method in the DSL is needed for this

[siddarthreddygsr]

While that works for predefined starting points, it doesn’t cover dynamic use cases like resuming arbitrary checkpoints. Currently, if a WebSocket connection breaks, I can only resume from the beginning of the workflow or by routing through conditional edges (which would require making every node a start node—not practical). With a start_from_node method, I could resume execution exactly where the interruption occurred, using the saved node and state.

[siddarthreddygsr]

I am willing on work on this feature.