netzSINN GPT Plugin

About

Features

• Interchangeable AI providers: OpenAI, Mistral AI (EU servers, GDPR-compliant), Anthropic, Google, Groq, DeepSeek, OpenRouter, Custom (i. e. Ollama or any other). And for translation DeepL (EU servers, GDPR-compliant).
• Admin-configurable AI contexts in the backend: context key + system prompt + individual parameters
• Event-based context system: other plugins can register their own contexts server-side
• Two usage paths: HTTP endpoint (AJAX, frontend JS) and direct PHP method call (server-side)
• API key encryption: AES-256 with native PHP Sodium, using the Joomla site secret as the key
• Protection: Rate limiting, CSRF token verification, origin check, and input sanitisation by default
• Per-context parameters: max_tokens,max_input_length, andtemperature configurable per context
• Backend test system: verify the API connection directly inside the Plugin Manager
• Conversation history: supports modern object-array format (role/content) and legacy strings
• Stateless: no conversation data is stored; every request is independent
• Built-in DeepL translation: translate text via HTTP or PHP with the same key management, CSRF protection, and rate limiting as the AI feature
• DeepL context system: define named translation profiles (e.g. `html_de`, `formal_en`) in the backend — no per-call parameter juggling needed

GDPR & Security

How it works

Explicit request parameter  >  Context value  >  Global admin default

For Developers

const reply = await NetzsinnGptAPI.ask({
    message: 'What are your opening hours?',
    context: 'faq',
    token: Joomla.getOptions('csrf.token')
});
document.getElementById('answer').textContent = reply;

<?php
$gpt = $this->getApplication()->bootPlugin('netzsinn_gpt', 'system');

$reply = $gpt->ask(
    message: 'What are your opening hours?',
    context: 'faq',
    history: [] // optional: previous conversation turns
);

<?php
public function onNetzsinnGptRegisterContexts(RegisterContextsEvent $event): void
{
    $event->addContexts([
        'tracking_report' => [
            'system_prompt'    => 'You are a data analysis assistant...',
            'max_input_length' => 5000,
            'max_tokens'       => 1200,
            'temperature'      => 0.3,
        ],
    ]);
}

// Simple — use global admin defaults
const translation = await NetzsinnGptAPI.translate({
    text:  'Hello, how are you?',
    token: Joomla.getOptions('csrf.token')
});

// With a named context (profile stored in backend)
const translation = await NetzsinnGptAPI.translate({
    text:    '<p>Hello World</p>',
    context: 'html_de',
    token:   Joomla.getOptions('csrf.token')
});

// Context + explicit override (override wins)
const translation = await NetzsinnGptAPI.translate({
    text:        '<p>Hello World</p>',
    context:     'html_de',
    tagHandling: 'text',   // overrides the context value
    token:       Joomla.getOptions('csrf.token')
});

$gpt = $this->getApplication()->bootPlugin('netzsinn_gpt', 'system');

// Simple — global admin defaults apply
$translation = $gpt->translate(
    text:       'Hello, how are you?',
    targetLang: 'DE'
);

// All optional parameters
$translation = $gpt->translate(
    text:               '<p>Hello World</p>',
    targetLang:         'DE',
    sourceLang:         'EN',
    tagHandling:        'html',
    formality:          'more',
    modelType:          'quality_optimized',
    preserveFormatting: true,
    splitSentences:     '1'
);

use Netzsinn\Plugin\System\NetzsinnGpt\Event\RegisterDeepLContextsEvent;

public function onNetzsinnGptRegisterDeepLContexts(RegisterDeepLContextsEvent $event): void
{
    $event->addContexts([
        'html_de' => [
            'target_lang'  => 'DE',
            'tag_handling' => 'html',
            'formality'    => 'more',
            // omitted fields → null = global admin value as fallback
        ],
        'plain_en' => [
            'target_lang'  => 'EN-GB',
            'tag_handling' => 'text',
        ],
    ]);
}

System Requirements

• Joomla 6.0+
• PHP 8.1+
• PHP Sodium Extension (libsodium) — included by default in modern PHP installations
• A valid API key from OpenAI or Mistral AI

Changelog

• v1.11.0

  • Security: New API keys are now stored in a new ENC-prefixed hex format; older saved keys still work and can be migrated automatically
  • Security: Encryption checks and decryption now recognise the new format strictly while still supporting older saved keys
  • Fix: Direct error logging was removed from decryption; masked messages now only go through the plugin log
• v1.10.3

  • Fix: API test popup now only shows contexts that are actually configured; removed the static faq and contact options; falls back to example or the first available context
  • Docs: Documentation updated to use example as the default context and explain configuration-based context selection in the test popup
• v1.10.2

  • refactor: package cleanup
• v1.10.1

  • Style: GPL license notice ensured in all PHP files (tests/, test/)
  • Style: Security guard placement verified in all src/ files (after namespace, before use)
• v1.10.0

  • Improved privacy warning hints — China providers (DeepSeek, Qwen) upgraded to alert-danger (red)
  • OpenRouter warning extended to note possible data transfer to China depending on selected model
  • Feature: Alibaba Qwen support — Qwen models (qwen-max, qwen-plus, qwen-turbo and more) are now available as an AI provider; includes a GDPR notice (Chinese service)
  • Feature: xAI Grok support — Grok models (grok-3, grok-3-fast, grok-3-mini, grok-2) are now available as an AI provider
  • Feature: Both new providers can be configured in the backend with their own API key and model selection, and are available as per-context provider overrides
    
• v1.9.0

  • Feature: Dynamic model list — available AI models are no longer hardcoded; a "Reload models" button fetches the current list directly from the provider's API
  • Feature: All model defaults and fallback lists are now managed centrally in one place, making future updates easier
  • Fix: CSRF token handling and response parsing for the model reload corrected
• v1.8.1

  • Fix: API test button now works correctly for all providers (DeepSeek, Groq, Anthropic, Google, OpenRouter, Custom) — previously only OpenAI and Mistral were supported
  • Feature: HTTP timeout configurable in the backend (10–300 seconds, default 120 s) under "Security & Debug"
  • Improvement: Default contexts reduced to a single generic example entry
• v1.8.0

  • Feature: DeepL context system — translation settings (target language, formality, etc.) can now be configured per context key, just like AI contexts
  • Feature: Third-party plugins can register their own DeepL contexts via a new event
  • Feature: Per-context translation parameters — admin settings always take priority; plugin-registered values only apply where the admin has not set a value
  • Feature: The translate() API call now accepts an optional context key; individual per-request parameters still override the context values
• v1.7.1

  • Fix: Unexpected PHP errors during AJAX requests are now correctly caught and logged instead of failing silently
  • Fix: Missing or invalid DeepL API responses now produce a clear error message instead of a silent crash
  • Improvement: DeepL translation requests are now logged in detail (text length, target language, parameters, result, errors)
  • Improvement: Debug & Diagnostics settings (rate limit, debug mode, API test button) are now grouped in a dedicated backend tab
• v1.7.0

  • Feature: Additional DeepL translation options — formality, model type, formatting preservation, and sentence splitting can now be configured in the backend as default values
  • Feature: DeepL API test button in the backend — opens a modal to test the translation directly with the configured settings
• v1.6.0

  • Feature: DeepL translation — the plugin can now translate texts directly via the DeepL API, independently of the AI chat providers
  • Feature: DeepL settings in the backend — API key, default target language (30 languages), optional source language, and GDPR notice (DeepL is an EU/German service)
  • Security: API key storage improved — keys are now saved with a clear prefix to prevent incorrect detection of already-encrypted values
  • Security: Old API keys in the previous format are automatically migrated to the new format on the next backend visit</item>
• v1.5.1

  • Fix: Admin settings (max. tokens, temperature, etc.) now always take priority over values set by other plugins — only the system prompt can still be overridden by other plugins
• v1.5.0

  • Feature: OpenRouter support — access to 200+ AI models (GPT-4o, Claude, Llama, Gemini, DeepSeek and more) with a single API key
  • Feature: Per-context provider and model — each AI context in the backend can now use a different provider and model
  • Fix: API keys for all providers are now correctly encrypted when saving
• v1.4.1

  • Improvement: The backend now has a single Test button; the provider is selected via a dropdown instead of separate buttons
    
  • Fix: Test button now correctly sends the request to the selected provider
  • Improvement: Label of the Debug Mode button revised
• v1.4.0

  • Feature: Multi-provider support — DeepSeek, Groq (Llama), Anthropic Claude, Google Gemini, and custom OpenAI-compatible services
  • Feature: GDPR notices for all non-EU providers (DeepSeek from China, Groq/Anthropic/Google from the US)
  • Improvement: Internal code refactoring for better maintainability and automated testing
• v1.3.1

  • Fix: Event-registered contexts without an explicit max_tokens now correctly fall back to the global admin value (instead of hardcoded 400)
  • Fix: Admin panel contexts now take priority over event-registered contexts with the same key

• v1.3.0 

  • Feature: Per-context parameters — max_tokens, max_input_length, temperature configurable per context
  • Feature: max_tokens moved from global-only to per-context (global value remains as fallback)
  • Feature: Subform layout changed to vertical blocks (no more horizontal scrolling)
  • Feature: Native support for modern history format (object-array with role/content)
  • Refactor: Internal context data structure changed from string to array-of-arrays

• v1.2.3 

  • Fix: loadCustomContexts() correctly reads subform data
  • Feature: Custom RegisterContextsEvent class for a stable event system

• v1.2.2 

  • Security: API keys in logs are now masked (first 10 + last 4 characters)

• v1.2.1 

  • Feature: AES-256 API key encryption with PHP Sodium
• v1.0.0 First official release

License

Download