How to output tool use chunk to the front end user?

[bmc0923]

I would like to output a chunk to the user every time a tool is called. This is gives instant feedback and helps the user have a good understanding of what's happening. I did try simply outputting a chunk in the tool (which is problematic when trying to keep it separate from front end code also), but it only outputs when the LLM has finished processing (not using stream currently). Any ideas?

Overall, I would just like to better understand how tools are being called in Neuron AI, because it's quite abstract and I'm struggling to quickly interact with it without digging deeper. Would love ideas. Thanks!

[bmc0923]

Dropping this into my agent class works. Not sure if this is a good solution?

public function chat(array|Message $messages): Message
  {
    // Normalize to array for parent::chat()
    $msgs = is_array($messages) ? $messages : [$messages];

    // Attach observer on "tool-calling"
    $this->observe(new class implements SplObserver {
      public function update(SplSubject $subject, $event = null, $data = null): void
      {
        if ($event === 'tool-calling' && $data instanceof ToolCalling) {
          // access the tool via the public property
          $tool = $data->tool;

          echo "data: " . json_encode(['type' => 'progress', 'text' => "[🔧 Tool Calling - " . $tool->getName() . " ] \n"]) . "\n\n";
          // flush PHP buffers immediately
          if (ob_get_length() !== false) {
            ob_flush();
          }
          flush();
        }
      }
    }, 'tool-calling');

    // Delegate to the core chat implementation
    return parent::chat($msgs);
  }