Struct Agent 

pub struct Agent<C: Client> { /* private fields */ }

Agent that automatically executes tools in a loop.

Unlike the raw Client, an Agent handles tool execution automatically:

1. Sends request with tool definitions from the MCP server (if configured)
2. Receives response with potential function calls
3. Executes tools automatically
4. Adds results back to conversation
5. Loops until no more function calls

Example

use unia::agent::Agent;
use unia::providers::{Gemini, Provider};
use unia::model::{Message, Part};

let client = Gemini::create("api_key".to_string(), "gemini-3.0-pro".to_string());
let agent = Agent::new(client)
    .with_server(weather_server);

let messages = vec![
    Message::User(vec![
        Part::Text { content: "What's the weather?".into(), finished: true }
    ])
];

let response = agent.chat(messages).await?;

Implementations

impl<C: Client> Agent<C>

pub fn new(client: C) -> Self

Create a new agent.

Arguments

• client: The initialized client instance

Tools are fetched from the configured MCP server if available.

pub fn with_server<S: MCPServer + 'static>(self, server: S) -> Self

Set the MCP server for the agent.

pub fn with_max_iterations(self, max: usize) -> Self

Set the maximum number of iterations for the agentic loop.

pub async fn chat( &self, messages: Vec<Message>, ) -> Result<Response, ClientError>

Send a chat request with automatic tool execution.

This method automatically handles the tool execution loop:

• Sends request to LLM with tools from the MCP server (if configured)
• Executes any tool calls
• Continues until no more tool calls or max iterations reached

Arguments

• messages: Conversation messages

Returns

The response containing all new messages generated during the execution (including tool calls and results)

pub fn chat_stream<'a>( &'a self, messages: Vec<Message>, ) -> Pin<Box<dyn Stream<Item = Result<Response, ClientError>> + Send + 'a>>

where C: StreamingClient,

Send a streaming chat request with automatic tool execution.

This method automatically handles the tool execution loop with streaming:

• Sends streaming request to LLM with tools from the MCP server (if configured)
• Executes any tool calls
• Continues until no more tool calls or max iterations reached

Arguments

• messages: Conversation messages

Returns

A stream of chunks for the final response after all tool executions complete