Struct LoopConfig 

pub struct LoopConfig<'a> {
    pub model: String,
    pub system_prompt: String,
    pub tools: Vec<ToolDef>,
    pub agent_tools: &'a [Box<dyn AgentTool>],
    pub extensions: &'a [Box<dyn Extension>],
    pub tool_execution: ToolExecutionMode,
    pub steering_queue: Option<&'a Mutex<PendingMessageQueue>>,
    pub follow_up_queue: Option<&'a Mutex<PendingMessageQueue>>,
    pub transform_context: Option<TransformFn>,
    pub prepare_next_turn: Option<PrepareNextTurnFn>,
    pub should_stop_after_turn: Option<ShouldStopFn>,
}

Configuration for the agent loop.

Fields

model: String

system_prompt: String

tools: Vec<ToolDef>

agent_tools: &'a [Box<dyn AgentTool>]

extensions: &'a [Box<dyn Extension>]

tool_execution: ToolExecutionMode

Tool execution mode: parallel (default) or sequential.

steering_queue: Option<&'a Mutex<PendingMessageQueue>>

Optional steering queue: messages delivered after the current assistant turn’s tool calls finish, before the next LLM call.

follow_up_queue: Option<&'a Mutex<PendingMessageQueue>>

Optional follow-up queue: messages delivered only after the agent has no more tool calls (fully idle).

transform_context: Option<TransformFn>

Optional transform applied to the message list before each LLM call. Receives the current messages and returns (possibly modified) messages. Pi-compatible: transformContext for context window management, pruning, etc.

prepare_next_turn: Option<PrepareNextTurnFn>

Optional callback invoked after each turn completes. Can return a TurnUpdate to modify the context for the next turn. Pi-compatible: prepareNextTurn.

should_stop_after_turn: Option<ShouldStopFn>

Optional predicate invoked after each turn completes. Return true to stop the agent loop early. Pi-compatible: shouldStopAfterTurn.