Struct AgentRuntime 

pub struct AgentRuntime<T: State> { /* private fields */ }

The core execution engine for agent graphs

Responsible for:

• Managing the graph structure
• Executing nodes in order
• Applying edge conditions
• Maintaining service clients
• Executing middleware hooks

Implementations

impl<T: State> AgentRuntime<T>

pub fn from_config(config: AgentConfig) -> Result<Self>

Create a new runtime from a configuration

This performs:

• Configuration validation
• LLM client initialization
• MCP client setup
• Graph construction

pub fn register_node( &mut self, name: &str, function: NodeFunction<T>, ) -> Result<()>

Register a handler function for a node

Replaces the placeholder function with the actual handler. Called during agent initialization.

pub fn register_edge_condition( &mut self, from: &str, condition_name: &str, condition_fn: EdgeCondition<T>, ) -> Result<()>

Register a condition function for an edge

Associates a condition function with conditional edges.

pub fn set_checkpointer( &mut self, checkpointer: Arc<MemoryCheckpointer>, ) -> &mut Self

Set the checkpointer for thread-scoped state persistence.

When set, use execute_with_thread(thread_id, state) to load/save state per thread.

pub fn with_checkpointer(self, checkpointer: Arc<MemoryCheckpointer>) -> Self

Builder-style: set the checkpointer.

pub fn checkpointer(&self) -> Option<Arc<MemoryCheckpointer>>

Get the checkpointer if set.

pub fn add_middleware( &mut self, middleware: Arc<dyn Middleware<T>>, ) -> &mut Self

Add middleware to the middleware pipeline

Middleware will be executed in the order they are added during node execution. Each middleware receives an ExecutionContext and can:

• Log/monitor execution
• Validate state
• Modify state before/after execution
• Skip nodes or abort execution

Arguments

• middleware - The middleware to add (must implement Middleware trait)

Example

runtime.add_middleware(Arc::new(LoggingMiddleware::new()));

pub fn middleware_pipeline(&self) -> &MiddlewarePipeline<T>

Get a reference to the middleware pipeline

pub fn middleware_pipeline_mut(&mut self) -> &mut MiddlewarePipeline<T>

Get mutable reference to the middleware pipeline

pub fn with_observability(&mut self) -> &mut Self

Add observability middleware (execution tracing, node timing, token usage, failure snapshots).

pub async fn execute(&self, initial_state: T) -> Result<T>

Execute the agent graph from start to end.

For thread-scoped persistence (resume, multi-turn), use execute_with_thread and set a checkpointer via set_checkpointer or config.

pub async fn execute_with_thread( &self, thread_id: &str, initial_state: T, ) -> Result<T>

Execute with a thread id for checkpointing. When a checkpointer is set, state is loaded from the last checkpoint for this thread (if any) and saved after each node.

pub fn graph(&self) -> &Graph<T>

Get a reference to the graph for inspection

pub fn graph_mut(&mut self) -> &mut Graph<T>

Get mutable access to the graph

pub fn config(&self) -> &AgentConfig

Get the configuration

pub fn llm_client(&self) -> Arc<dyn LLMClient>

Get the LLM client

pub fn mcp_client(&self, name: &str) -> Option<Arc<dyn MCPClient>>

Get an MCP client by name

pub fn visualize_graph(&self, output_path: &str) -> Result<()>

Visualize the agent graph as a text file

Generates a text-based representation of the graph structure. This is useful for debugging and documentation.

Arguments

• output_path: Path where to save the visualization

Example

#[cfg(feature = "visualization")]
runtime.visualize_graph("agent_graph.txt")?;