Struct AgentActivities 

pub struct AgentActivities {
    pub llm: Arc<dyn LLMProvider>,
    pub tools: ToolRegistry,
}

Shared state for the activity worker. One instance per worker process.

Arc<dyn LLMProvider> and the ToolRegistry are both cheap to clone and safe to share across concurrent activity executions.

Fields

llm: Arc<dyn LLMProvider>

tools: ToolRegistry

Implementations

impl AgentActivities

pub fn new(llm: Arc<dyn LLMProvider>, tools: ToolRegistry) -> Self

impl AgentActivities

pub async fn __llm_chat( self: Arc<Self>, _ctx: ActivityContext, input: LlmChatInput, ) -> Result<LlmResponse, ActivityError>

One LLM “reasoning step”: given the running conversation and the catalog of tools, ask the model what to do next.

Returns either a final answer or a list of tool calls to execute.

pub async fn __execute_tool( self: Arc<Self>, _ctx: ActivityContext, call: ToolCall, ) -> Result<ToolResult, ActivityError>

Execute a single tool call.

Tool-side failures are returned as Ok(ToolResult { error: Some(..) }) — they are observed by the LLM, not retried by Temporal. Only infrastructure errors (missing tool registration, serde failure) surface as Err.

impl AgentActivities

pub const llm_chat: LlmChat = agent_activities::LlmChat

pub const execute_tool: ExecuteTool = agent_activities::ExecuteTool

Trait Implementations

impl Clone for AgentActivities

fn clone(&self) -> AgentActivities

Returns a duplicate of the value.

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

Performs copy-assignment from source.