Struct AgentRuntime 

pub struct AgentRuntime { /* private fields */ }

Implementations

impl AgentRuntime

pub async fn create_session(&self) -> SessionId

pub async fn restore_session( &self, session_id: &SessionId, ) -> Option<AgentSession>

pub async fn session(&self, session_id: &SessionId) -> Option<AgentSession>

pub async fn session_or_err( &self, session_id: &SessionId, ) -> AgentResult<AgentSession>

pub async fn with_session_mut<F, R>( &self, session_id: &SessionId, f: F, ) -> AgentResult<R>

where F: FnOnce(&mut AgentSession) -> R,

pub fn emit_event(&self, event: RuntimeEvent)

pub fn subscribe_runtime_events(&self) -> Receiver<RuntimeEvent>

Subscribe to runtime events from the internal broadcast channel.

Events are delivered directly from the runtime’s event bus (capacity 2048). Slow consumers may receive Lagged(n) errors if they cannot keep up — ensure the receiver loop processes events promptly or use a buffering layer in the consumer if backpressure is a concern.

pub fn session_manager(&self) -> &SessionManager

pub fn llm_engine(&self) -> &LlmEngine

pub fn provider(&self) -> Arc<dyn LlmProvider>

pub fn set_client(&mut self, provider: Arc<dyn LlmProvider>)

Replace the LLM provider at runtime (e.g., model switch). Requires &mut self — obtain via runtime.lock().await.

pub fn tools_mut(&self) -> Arc<RwLock<ToolRegistry>>

pub fn config(&self) -> RwLockReadGuard<'_, AgentConfig>

pub async fn set_reasoning_effort(&self, effort: ReasoningEffort)

设置 reasoning effort（异步版本）

pub fn set_reasoning_effort_sync(&self, effort: ReasoningEffort)

设置 reasoning effort（同步版本，只在同步上下文中使用）

pub fn approval_handler(&self) -> Option<&Arc<dyn ApprovalHandler>>

pub fn tool_policy(&self) -> Option<&Arc<dyn ToolPolicy>>

pub async fn cached_approval( &self, session_id: &SessionId, action_key: &str, ) -> bool

pub async fn cache_approval(&self, session_id: &SessionId, action_key: String)

pub async fn save_checkpoint( &self, session_id: &SessionId, checkpoint: CheckpointData, ) -> AgentResult<()>

pub async fn load_checkpoint( &self, _session_id: &SessionId, _checkpoint: &CheckpointData, ) -> AgentResult<Option<CheckpointData>>

pub async fn run<F>( &self, session_id: SessionId, on_event: F, ) -> AgentResult<RunOutcome>

where F: FnMut(RuntimeEvent) -> AgentResult<()> + Send + 'static,

pub async fn run_turn<F>( &self, session_id: SessionId, user_input: &str, on_event: F, ) -> AgentResult<RunOutcome>

where F: FnMut(RuntimeEvent) -> AgentResult<()> + Send + 'static,

pub async fn run_turn_collect( &self, session_id: SessionId, user_input: &str, ) -> AgentResult<(Vec<RuntimeEvent>, RunOutcome)>

pub async fn add_user_message( &self, session_id: &SessionId, text: impl Into<String>, ) -> AgentResult<()>

pub async fn add_system_message( &self, session_id: &SessionId, text: impl Into<String>, ) -> AgentResult<()>

pub async fn add_tool_result( &self, session_id: &SessionId, tool_call_id: &str, summary: impl Into<String>, ) -> AgentResult<()>

pub async fn get_messages( &self, session_id: &SessionId, ) -> AgentResult<Vec<ChatMessage>>

pub async fn set_messages( &self, session_id: &SessionId, messages: Vec<ChatMessage>, ) -> AgentResult<()>

Replace the chat messages for a session — only for persistence restore. Validates message sequence before applying.

仅供持久化恢复使用。

pub async fn validate_session(&self, session_id: &SessionId) -> AgentResult<()>

pub fn session_store(&self) -> Arc<dyn SessionStore>

pub fn on_turn_end<F>(&self, f: F)

where F: Fn(&TurnContext) + Send + Sync + 'static,

Register a turn-end callback. The callback receives a TurnContext with raw data about the completed turn iteration. Consumers (e.g. phi-telemetry) use this to build their own metrics without agent-base knowing anything about metrics.

pub fn cancel(&self)

Cancel the currently executing run_turn / run. No-op if there is no current execution.

pub fn reset_cancel(&self)

Reset the cancel token (called automatically before each run_turn)

pub fn cancel_token(&self) -> CancellationToken

Get a clone of the cancel token

pub fn is_cancelled(&self) -> bool

Check if cancellation has been requested

pub fn steer(&self, message: String)

Push a steering message — will be processed at the start of the next turn in the current run_managed() loop.

pub fn follow_up(&self, message: String)

Push a follow-up message — will be processed after the inner turn loop stops naturally (no tool calls or max turns).

pub async fn run_managed<F>( &self, session_id: SessionId, user_input: &str, on_event: F, ) -> AgentResult<RunOutcome>

where F: FnMut(RuntimeEvent) -> AgentResult<()> + Send + 'static,

Run the agent in managed mode with message queue support.

This wraps run_turn() (or run()) in an outer loop: after the inner turn loop completes, any follow-up messages are drained and a new inner loop is started. Steering messages are drained automatically at each iteration of the inner turn loop.

The on_event callback receives all events from every inner run.

pub fn set_queue_mode(&self, mode: QueueMode)

Set the drain mode for the message queues.

Trait Implementations

impl Clone for AgentRuntime

fn clone(&self) -> AgentRuntime

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.