Struct AgentHooks 

pub struct AgentHooks {
    pub should_stop_after_turn: Option<Arc<dyn Fn(&ShouldStopAfterTurnContext) -> bool + Send + Sync>>,
    pub before_tool_call: Option<Box<dyn Fn(&BeforeToolCallContext) -> BeforeToolCallResult + Send + Sync>>,
    pub after_tool_call: Option<Box<dyn Fn(&AfterToolCallContext) -> AfterToolCallResult + Send + Sync>>,
    pub get_steering_messages: Option<Arc<dyn Fn() -> Vec<Message> + Send + Sync>>,
    pub get_follow_up_messages: Option<Arc<dyn Fn() -> Vec<Message> + Send + Sync>>,
    pub tool_execution: ToolExecutionMode,
}

Stable since 0.63.0

Callback hooks for the agent loop.

These mirror pi-mono’s AgentLoopConfig hooks, allowing callers to inject custom logic at key points in the agentic loop.

Fields

should_stop_after_turn: Option<Arc<dyn Fn(&ShouldStopAfterTurnContext) -> bool + Send + Sync>>

Called after each turn completes. Return true to stop the agent loop.

Wrapped in Arc so the hook can be invoked multiple times without being consumed (unlike Box<dyn Fn> which requires take()).

before_tool_call: Option<Box<dyn Fn(&BeforeToolCallContext) -> BeforeToolCallResult + Send + Sync>>

Called before a tool is executed. Return a BeforeToolCallResult with block: true to prevent execution.

after_tool_call: Option<Box<dyn Fn(&AfterToolCallContext) -> AfterToolCallResult + Send + Sync>>

Called after a tool execution completes. Can override the result.

get_steering_messages: Option<Arc<dyn Fn() -> Vec<Message> + Send + Sync>>

Returns steering messages to inject mid-run. Called after each turn (unless stopped).

get_follow_up_messages: Option<Arc<dyn Fn() -> Vec<Message> + Send + Sync>>

Returns follow-up messages to process after the agent would stop. Called when the agent has no more tool calls and no steering messages.

tool_execution: ToolExecutionMode

Tool execution mode.

Trait Implementations

impl Default for AgentHooks

fn default() -> AgentHooks

Returns the “default value” for a type.