Struct LlmAgent 

pub struct LlmAgent { /* private fields */ }

An agent that uses an LLM (Large Language Model) to process messages.

The LlmAgent integrates with the Ceylon Mesh agent system and delegates message processing to an LLM. It supports:

• Configurable system prompts
• Model parameters (temperature, max tokens, etc.)
• Tool calling integration with existing ToolInvoker
• Multiple LLM providers (OpenAI, Anthropic, Ollama, etc.)
• Optional memory module integration

Examples

use runtime::llm::{LlmAgent, LLMConfig};

// Create agent with Ollama (local, no API key needed)
let agent = LlmAgent::builder("my_agent", "ollama::llama2")
    .with_system_prompt("You are a helpful assistant.")
    .build()
    .expect("Failed to create agent");

// Create agent with OpenAI
let agent = LlmAgent::builder("gpt_agent", "openai::gpt-4")
    .with_api_key(std::env::var("OPENAI_API_KEY").unwrap())
    .with_temperature(0.7)
    .build()
    .expect("Failed to create agent");

Implementations

impl LlmAgent

pub fn builder( name: impl Into<String>, model: impl Into<String>, ) -> LlmAgentBuilder

Create a builder for constructing an LlmAgent.

Arguments

• name - The name of the agent
• model - The model in “provider::model” format (e.g., “openai::gpt-4”, “ollama::llama2”)

Examples

use runtime::llm::LlmAgent;

let agent = LlmAgent::builder("my_agent", "ollama::llama2")
    .build()
    .expect("Failed to create agent");

pub fn new_with_config( name: impl Into<String>, config: LLMConfig, system_prompt: impl Into<String>, memory: Option<Arc<dyn Memory>>, ) -> Result<Self>

Create an LlmAgent with comprehensive LLMConfig

pub fn with_react(&mut self, config: ReActConfig)

Enable ReAct (Reason + Act) mode

pub async fn send_message_react( &mut self, message: impl Into<String>, ctx: &mut AgentContext, ) -> Result<ReActResult>

Send a message using ReAct reasoning mode

pub async fn send_message_and_get_response( &mut self, message: impl Into<String>, ctx: &mut AgentContext, ) -> Result<String>

Send a message and get the LLM’s response This is a convenience method for Python bindings and direct usage. It processes the message with the LLM and returns the response text.

pub fn last_response(&self) -> Option<String>

Get the last assistant response from conversation history

Trait Implementations

impl Agent for LlmAgent

fn name(&self) -> String

fn on_message<'life0, 'life1, 'async_trait>( &'life0 mut self, msg: CeylonMessage, ctx: &'life1 mut AgentContext, ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

fn on_generic_message<'life0, 'life1, 'async_trait>( &'life0 mut self, msg: GenericMessage, ctx: &'life1 mut AgentContext, ) -> Pin<Box<dyn Future<Output = Result<GenericResponse>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Handle a generic string message and return a generic response. Default implementation returns an echo response.

fn tool_invoker(&self) -> Option<&ToolInvoker>

Get the tool invoker for this agent (if it has actions)

fn tool_invoker_mut(&mut self) -> Option<&mut ToolInvoker>

Get mutable tool invoker for registration

fn on_start<'life0, 'life1, 'async_trait>( &'life0 mut self, _ctx: &'life1 mut AgentContext, ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

fn on_stop<'life0, 'life1, 'async_trait>( &'life0 mut self, _ctx: &'life1 mut AgentContext, ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,