Version 1.3

Changes summary

• Added human message role for human-in-the-loop scenarios
• Triggering custom callbacks after remote message is received. Usefull for hooking thread moderation functions.
• Dynamic enabling and disabling tools. For scenarios where tools must be temporary disabled while executing the tread. Example: when prompt injection is detected.
• Method for retrieving last tool result.
• Added executionTime stat to Thread usage.
• Improved API communication
• Added new models
  
  
  

Triggering message callbacks

The code bellow will trigger custom callback function everytime message is added to the thread.

$thread->OnMessage(function($e) {   
            $thread = $e->thread ?? null;
            $message = $e->message ?? null;
            $usage = $e->usage ?? null;
            $model = $e->model ?? $thread->GetModel();
});

// This will trigger the callback above
$thread->AddMessage(new \AIpi\Message('Hi!'));

// This will NOT trigger the callback above
$thread->AddMessage(new \AIpi\Message('Hi!'), ['runCallbacks' => false]);

Enable/disable tools

// Temporary disable tools
$thread->DisableTools();

// ...some thread execution

// Enable tools
$thread->EnableTools();

Retrieving last result from tools call

$toolResult = $thread->GetLastToolResult('SubmitEvaluation') ?? null;

New models

API communication is improved plus some new models to the list:

• google-gemini-2.5-pro
• google-gemini-2.5-flash
• openai-gpt-4.5-preview
• openai-gpt-3.5-turbo // old but gold