neuron-debugger

Neuron AI Debugger

This skill helps you debug and monitor Neuron AI applications using Inspector APM and the framework's observability system.

Inspector APM Integration

Inspector provides deep insights into agent execution, helping you understand:

• Why the model took certain decisions
• What data the model reacted to
• Timeline of all operations (LLM calls, tool usage, vector search)
• Performance bottlenecks and latency

Setup

1. Get an Inspector account at https://inspector.dev
2. Set the ingestion key in your environment:

# .env file
INSPECTOR_INGESTION_KEY=your_ingestion_key_here

1. Agent execution is automatically tracked - no code changes needed!

Viewing Execution Timeline

After running an agent, visit your Inspector dashboard to see:

[AI Inference] → [Tool Call: search_database] → [AI Inference] → [Response]
     ↓                    ↓                        ↓
  850ms               1.2s                     920ms

Each segment shows:

• Duration and timing
• Input/output data
• Errors if any occurred
• Metadata (model used, tokens, etc.)

Event System Observability

Neuron uses a static EventBus for monitoring all framework events.

Event Emission

All components emit events automatically:

use NeuronAI\Observability\EventBus;

// Workflows emit events like:
// - 'workflow-start'
// - 'workflow-resume'
// - 'workflow-end'
// - 'workflow-node-start'
// - 'workflow-node-end'
// - 'middleware-before-start'
// - 'middleware-after-end'
// - 'tool-calling'
// - 'tool-called'
// - 'rag-retrieving'
// - 'rag-retrieved'
// - 'inference-start'
// - 'inference-end'

Custom Observers

use NeuronAI\Observability\ObserverInterface;

class CustomObserver implements ObserverInterface
{
    public function handle(string $event, object $source, mixed $data): void
    {
        // Handle events
        echo "Event: {$event}\n";
        echo "Source: " . $source::class . "\n";
        var_dump($data);
    }
}