README

AIForm is a component for collecting structured data through multi-turn natural language conversations. It uses an AI agent to progressively gather information defined by a structured output class, validating each piece of data along the way.

The form maintains conversation history, tracks collected fields, missing fields, and validation errors across multiple turns.

Read the full article here: https://inspector.dev/collecting-data-through-conversation-introducing-aiform-for-neuronai/

Before moving on, support the Neuron AI community giving a GitHub star ⭐️. Thank you!

Creating a Data Class

You can provide the form structure you want to collect defining a data class using PHP attributes in the same way as you would with any other structured output class in Neuron AI.

You can also attach validation rules like #[NotBlank] or #[Email] to fields, to enforce data requirements.

use NeuronAI\StructuredOutput\SchemaProperty;
use NeuronAI\StructuredOutput\Validation\Rules\Email;
use NeuronAI\StructuredOutput\Validation\Rules\NotBlank;

class RegistrationData
{
    #[SchemaProperty(description: 'User full name', required: true)]
    #[NotBlank]
    public string $name;

    #[SchemaProperty(description: 'Email address', required: true)]
    #[Email]
    public string $email;

    #[SchemaProperty(description: 'Phone number')]
    public ?string $phone = null;

    #[SchemaProperty(description: 'Company name')]
    public ?string $company = null;
}

Creating a Custom Form

Extend the AIForm class to create your custom form:

<?php

namespace App\Neuron\Forms;

use NeuronAI\Form\AIForm;
use NeuronAI\Form\Enums\FormStatus;
use NeuronAI\Form\FormState;
use NeuronAI\Providers\Anthropic\Anthropic;

class RegistrationForm extends AIForm
{
    protected string $formDataClass = RegistrationData::class;

    protected function provider(): AIProviderInterface
    {
        return new Anthropic(
            key: 'ANTHROPIC_API_KEY',
            model: 'ANTHROPIC_MODEL',
        );
    }

    /**
     * Handle the submitted form data.
     */
    protected function callback(): mixed
    {
        // $data is an instance of RegistrationData
        // Save to database, send email, call API, etc.
        return function (RegistrationData $data) {
            $this->userService->register($data);
        };
    }
}

Using Form in Application Code

Basic Usage

use App\Forms\RegistrationForm;
use NeuronAI\Chat\Messages\UserMessage;
use NeuronAI\Providers\OpenAI\OpenAI;

// Create form instance with AI provider
$form = RegistrationForm::make();

// Turn 1
$state = $form->process(new UserMessage("Hi, I'd like to register"));
$state = $handler->run();

echo $state->getStatus()->value;        // 'incomplete'
echo $handler->getLastResponse();        // AI asks for name

// Turn 2
$handler = $form->process(new UserMessage("My name is John Doe"));
$state = $handler->run();

// Turn 3
$handler = $form->process(new UserMessage("john@example.com"));
$state = $handler->run();

// Check progress
echo $state->getCompletionPercentage();  // e.g., 75
print_r($state->getMissingFields());     // ['phone', 'company']

Web Application Example (Controller)

In a web application, you need to use a persistent storage for the chat history to retain messages across multiple requests. You can use one of the Neuron AI chat history components to store messages in a database or file system.

use App\Forms\RegistrationForm;
use NeuronAI\Chat\History\FileChatHistory;
use NeuronAI\Chat\Messages\UserMessage;
use NeuronAI\Workflow\Persistence\FilePersistence;

class RegistrationController
{
    public function __invoke(Request $request): Response
    {
        $sessionId = $request->session()->getId();

        // Create form with persistence for multi-request handling
        $form = RegistrationForm::make()
            ->setChatHistory(new FileChatHistory("/tmp/chats/{$sessionId}"));

        // Send the next user message
        $handler = $form->process(
            new UserMessage($request->input('message'))
        );

        $state = $handler->run();

        return response()->json([
            'status' => $state->getStatus()->value,
            'message' => $handler->getLastResponse(),
            'completion' => $state->getCompletionPercentage(),
            'missing_fields' => $state->getMissingFields(),
            'is_complete' => $form->isComplete(),
        ]);
    }
}

Requiring Confirmation

When you need users to review and confirm their data before submission, enable confirmation mode:

class RegistrationForm extends AIForm
{
    protected string $formDataClass = RegistrationData::class;

    protected function provider(): AIProviderInterface
    {
        return new Anthropic('key', 'model');
    }
}

// Enable confirmation in your application code
$form = RegistrationForm::make()->requireConfirmation(true);

Interruption

When confirmation is enabled and all required fields are collected, the form throws a FormInterruptRequest instead of submitting immediately. This interrupts the workflow and gives you control over the confirmation flow:

use NeuronAI\Form\Interrupt\FormInterruptRequest;

$handler = $form->process(new UserMessage("john@example.com"));
$state = $handler->run();

// Check if the form is waiting for confirmation
if ($handler->getInterrupt() instanceof FormInterruptRequest) {
    $interrupt = $handler->getInterrupt();

    // The interrupt contains the collected data for review
    $data = $interrupt->getData();

    // The AI-generated confirmation message
    echo $handler->getLastResponse();
    // "I've collected the following information:
    //  Name: John Doe
    //  Email: john@example.com
    //  Is this correct?"

    // Store the interrupt (e.g., in session) to resume later
    $_SESSION['form_interrupt'] = serialize($interrupt);
}

Resuming After Confirmation

To resume the form after the user responds to the confirmation prompt, pass the interrupt back to process():

// Retrieve the stored interrupt
$interrupt = unserialize($_SESSION['form_interrupt']);

// User confirms the data
$handler = $form->process(new UserMessage("Yes, that's correct"), $interrupt);
$state = $handler->run();

echo $state->getStatus()->value;  // 'complete'

// If the user wants to make changes
$handler = $form->process(new UserMessage("No, I need to change my email"), $interrupt);
$state = $handler->run();

echo $state->getStatus()->value;  // 'incomplete' - back to collecting data

Confirmation Phrases

The form recognizes these confirmation responses:

• Confirm: "yes", "confirm", "correct", "right", "ok", "okay", "sure", "yep", "yeah"
• Reject: "no", "cancel", "wrong", "incorrect", "change", "edit", "nope"

Handling Cancellation

Users can cancel the form at any time using exit phrases:

$handler = $form->process(new UserMessage("cancel"));
$state = $handler->run();

echo $state->getStatus()->value;  // 'closed'

Accessing Form Data

// Get collected data (available during collection)
$data = $form->getData();
// or
$data = $state->getCollectedData();

// Get submitted data (available after submission)
$submitted = $state->getSubmittedData();

// Get form state details
$missing = $state->getMissingFields();        // ['email', 'phone']
$errors = $state->getValidationErrors();      // ['name' => ['must not be blank']]
$completion = $state->getCompletionPercentage(); // 50

Custom Exit Detection

Override detectExit() for custom cancellation logic:

class SmartForm extends AIForm
{
    protected function detectExit(string $content): bool
    {
        // Add custom exit detection
        if (preg_match('/no\s+thanks|not\s+interested/i', $content)) {
            return true;
        }

        return parent::detectExit($content);
    }
}

Monitoring

Since this component is build on top of Neuron AI Workflow, it natively supports monitoring and debugging using the built-in Inspector integration.

After you sign up, make sure to set the INSPECTOR_INGESTION_KEY variable in the application environment file to start monitoring:

INSPECTOR_INGESTION_KEY=fwe45gtxxxxxxxxxxxxxxxxxxxxxxxxxxxx

After configuring the environment variable, you will see the agent execution timeline in your Inspector dashboard.

Learn more about Monitoring in the documentation.

Behind the scenes

This package is built on top of the Neuron AI Workflow component, just like the Agent class, but designed toward a specific use case: collecting structured data from the user conversation.

• Neuron AI Workflow: https://docs.neuron-ai.dev/workflow/getting-started
• Neuron AI Agent: https://docs.neuron-ai.dev/agent/agent
• Neuron AI Structured Output: https://docs.neuron-ai.dev/agent/structured-output

Contribution

You are welcome to contribute to this project by submitting pull requests or issues.