Skip to main content

​
Overview

The simplest use of an Agent - a straightforward Q&A without tools. The agent uses the LLM directly to answer questions. This demonstrates the core agent loop: receiving a message, processing it through the LLM, and returning a response. Key concepts:
  • AgentBuilder: Constructs configured agent instances
  • AgentState: Immutable state container for messages and metadata
  • Agent::finalStep(): Executes the agent loop until completion

​
Example

<?php
require 'examples/boot.php';

use Cognesy\Addons\Agent\AgentBuilder;
use Cognesy\Addons\Agent\Core\Data\AgentState;
use Cognesy\Messages\Messages;

// Build a basic agent
$agent = AgentBuilder::new()
    ->withLlmPreset('anthropic')  // Optional: specify LLM
    ->build();

// Create initial state with user question
$state = AgentState::empty()->withMessages(
    Messages::fromString('What is the capital of France? Answer in one sentence.')
);

// Execute agent until completion
$finalState = $agent->finalStep($state);

// Extract response
$response = $finalState->currentStep()?->outputMessages()->toString() ?? 'No response';

echo "Answer: {$response}\n";
echo "Steps: {$finalState->stepCount()}\n";
echo "Status: {$finalState->status()->value}\n";
?>

​
Expected Output

Answer: The capital of France is Paris.
Steps: 1
Status: finished

​
Key Points

  • No tools: This agent has no tools attached, so it only uses LLM reasoning
  • Single step: Simple Q&A typically completes in one step
  • Immutable state: Each step returns a new AgentState instance
  • Status tracking: Final status indicates success/failure/timeout