Struct AgentClient 

pub struct AgentClient<S> { /* private fields */ }

High-level A2A client for interacting with agents

This client wraps a Tower service and provides convenient methods for common A2A operations. The service is generic over any implementation that satisfies the Service trait bounds.

Example

use tower_a2a::prelude::*;

let url = "https://agent.example.com".parse().unwrap();
let mut client = A2AClientBuilder::new(url)
    .with_http()
    .build()?;

let message = Message::user("Hello, agent!");
let task = client.send_message(message).await?;
println!("Task created: {}", task.id);

Implementations

impl<S> AgentClient<S>

where S: Service<A2ARequest, Response = A2AResponse, Error = A2AError>,

pub fn new(service: S, config: ClientConfig) -> Self

Create a new agent client

Arguments

• service - The Tower service that handles requests
• config - Client configuration

pub fn config(&self) -> &ClientConfig

Get the client configuration

pub async fn send_message(&mut self, message: Message) -> Result<Task, A2AError>

Send a message to the agent and get a task

Arguments

• message - The message to send to the agent

Returns

A task representing the agent’s processing of the message

Errors

Returns an error if the message fails to send or the response is invalid

pub async fn send_message_streaming( &mut self, message: Message, ) -> Result<Task, A2AError>

Send a message with streaming enabled

Note: Streaming is not yet fully implemented

pub async fn send_message_in_context( &mut self, message: Message, context_id: String, ) -> Result<Task, A2AError>

Send a message in a specific context for multi-turn conversations

Arguments

• message - The message to send
• context_id - The context ID for grouping related messages

pub async fn get_task(&mut self, task_id: String) -> Result<Task, A2AError>

Get a task by ID

Arguments

• task_id - The unique identifier of the task to retrieve

Returns

The task with the specified ID

Errors

Returns A2AError::TaskNotFound if the task doesn’t exist

pub async fn list_tasks( &mut self, status: Option<TaskStatus>, limit: Option<u32>, ) -> Result<Vec<Task>, A2AError>

List tasks with optional filtering

Arguments

• status - Optional filter by task status
• limit - Maximum number of tasks to return (default: 100)

Returns

A vector of tasks matching the query

pub async fn list_all_tasks(&mut self) -> Result<Vec<Task>, A2AError>

List all tasks without filtering

pub async fn list_tasks_by_status( &mut self, status: TaskStatus, ) -> Result<Vec<Task>, A2AError>

List tasks with a specific status

pub async fn cancel_task(&mut self, task_id: String) -> Result<Task, A2AError>

Cancel a task by ID

Arguments

• task_id - The unique identifier of the task to cancel

Returns

The updated task with cancelled status

pub async fn discover(&mut self) -> Result<AgentCard, A2AError>

Discover agent capabilities by fetching the Agent Card

This retrieves the agent’s metadata from /.well-known/agent-card.json

Returns

The agent’s capability card

pub async fn poll_until_complete( &mut self, task_id: String, poll_interval_ms: u64, max_attempts: usize, ) -> Result<Task, A2AError>

Poll a task until it reaches a terminal state

This is a convenience method that repeatedly calls get_task until the task is completed, failed, cancelled, or rejected.

Arguments

• task_id - The task ID to poll
• poll_interval - How often to poll (in milliseconds)
• max_attempts - Maximum number of polling attempts (0 = unlimited)

Returns

The final task state