Struct ToolNode 

pub struct ToolNode<S: CoreState> { /* private fields */ }

Built-in graph node that handles tool execution.

Reads ToolCall entries from the last AiMessage in state.messages(). Executes all tool calls in parallel (tokio::spawn per call). Appends ToolMessage results to state via StateUpdate.

Error handling

When handle_tool_errors is true (default), tool failures become ToolMessages with error text — the LLM sees the error and can retry. When false, the first tool error propagates as NodeResult::Error.

Unknown tool names always produce an error ToolMessage regardless of the handle_tool_errors setting.

Example

let tool_node = ToolNode::<MyState>::new(Arc::new(registry));

graph.add_node("tools", tool_node);
graph.add_conditional_edge("chat", tools_condition::<MyState>);
graph.add_edge("tools", "chat");

Implementations

impl<S: CoreState> ToolNode<S>

where S::Update: CoreStateUpdate,

pub fn new(registry: Arc<ToolRegistry>) -> Self

Create a new ToolNode with the given registry. Defaults to handle_tool_errors = true.

pub fn with_error_handling(self, handle: bool) -> Self

Set whether tool errors become ToolMessages (true) or propagate (false).

pub fn with_interceptor( self, interceptor: impl ToolCallInterceptor + 'static, ) -> Self

Set a ToolCallInterceptor that runs before each tool execution.

Trait Implementations

impl<S: CoreState> Debug for ToolNode<S>

where S::Update: CoreStateUpdate,

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<S: CoreState> NodeFn<S> for ToolNode<S>

where S::Update: CoreStateUpdate,

fn call(&self, state: &S, ctx: &NodeContext) -> NodeFuture<S::Update>

Execute the node with a state snapshot and execution context.

fn name(&self) -> &str

Human-readable name for logging and debugging