elgentos / magento2-hyva-ai-loes
Loes.ai provider for the Hyvä AI framework: Dutch, EU-hosted open-weight LLMs via the OpenAI-compatible HostYourAI router.
Package info
github.com/elgentos/magento2-hyva-ai-loes
Type:magento2-module
pkg:composer/elgentos/magento2-hyva-ai-loes
Requires
- php: ^8.3
- hyva-themes/magento2-module-ai: ^1.1
- magento/framework: *
- magento/module-backend: *
- magento/module-config: >=101.0.0
Requires (Dev)
None
Suggests
None
Provides
None
Conflicts
None
Replaces
None
README
Adds Loes.ai as a provider to the Hyvä AI framework
(hyva-themes/magento2-module-ai), next to the stock OpenAI, Anthropic, Gemini and DeepL providers.
Loes is a Dutch AI assistant by HostYourAI B.V. (Groningen). Its models are open-weight, run on GPUs in European datacenters and are never trained on your data. That makes it a good fit for stores that want AI features (CMS translations, content generation) without sending content to a US cloud provider.
Requirements
- Magento 2.4.7 or newer, PHP 8.3 or newer
- Hyvä AI (
hyva-themes/magento2-module-ai^1.1, available to Hyvä license holders) - A HostYourAI router API key (starts with
hyai-): sign up
Installation
composer require elgentos/magento2-hyva-ai-loes bin/magento module:enable Elgentos_HyvaAiLoes bin/magento setup:upgrade
Configuration
Stores › Configuration › Hyvä AI Providers › General › Loes.ai Configuration
| Field | Description |
|---|---|
| API Key | Your HostYourAI router key (hyai-…). Stored encrypted. |
| Model | The Loes model used by default (see below). |
| API URL | OpenAI-compatible chat completions endpoint, https only. Default https://hostyourai.com/api/v1/chat/completions; only change it for a dedicated Loes deployment. |
| Test connection | Sends a one-word request with the saved key and model, and shows the reply. |
Temperature and max tokens are configured per feature in the Hyvä AI provider options, like the other providers. The request timeout and concurrency limit from Hyvä AI › Runtime Safeguards apply.
Models
| Model | Value |
|---|---|
| Loes | hyai/loes-large |
| Super Loes | hyai/super-loes |
| Loes EU (EuroLLM 22B) | hyai/loes-large-eu-eurollm-22b-2512 |
Any other model on the HostYourAI router can be added from your own module's di.xml:
<type name="Hyva\Ai\Model\ProviderResolver"> <arguments> <argument name="providerModels" xsi:type="array"> <item name="loes" xsi:type="array"> <item name="options" xsi:type="array"> <item name="deepseek-v4-flash" xsi:type="array"> <item name="value" xsi:type="string">deepseek-ai/DeepSeek-V4-Flash</item> <item name="label" xsi:type="string">DeepSeek V4 Flash (EU)</item> </item> </item> </item> </argument> </arguments> </type>
How it works
The HostYourAI router is OpenAI-compatible, so the provider posts standard chat completion requests
(model, messages, temperature, max_tokens) with a Bearer key and returns the assistant reply.
A leading <think>…</think> reasoning block, which Qwen-based models can emit, is stripped from the
reply. The provider registers itself under the id loes in the Hyvä AI provider pool, model resolver and
options schema pool.
Native tool calling (ToolCapableProviderInterface) is not implemented yet. The HostYourAI docs do not
cover it.
Running the tests
The unit tests need Magento's framework and hyva-themes/magento2-module-ai on the autoloader. From a
Magento project that has this module installed:
vendor/bin/phpunit vendor/elgentos/magento2-hyva-ai-loes/Test/Unit
License
OSL-3.0. Copyright © Elgentos.
Loes.ai and HostYourAI are trademarks of HostYourAI B.V. This module is not affiliated with or endorsed by HostYourAI B.V. or Hyvä Themes B.V.