Reworked the Hugging Face provider (#944)

app/MindWork AI Studio/wwwroot/changelog/v26.8.2.md

@@ -7,6 +7,8 @@
 - Added the IONOS AI Model Hub as a provider for chats and embeddings. It runs open-source models in Germany, is subject to the GDPR, and IONOS states that your data is not used for training.
 - Added LiteLLM as a new LLM provider for chats, embeddings, and speech-to-text. LiteLLM is an AI gateway you run yourself in front of models from many providers. Because your gateway decides where your data goes, you set its trust level yourself. Thanks Prodman Devokadev (`prodmanpd`) for this first contribution.
 - Added speech-to-text for Helmholtz Blablador and GroqCloud, and embeddings for GWDG SAIA. These providers offer these services now, so you can select them when you dictate a message or when you set up a data source.
+- Added embeddings and speech-to-text for Hugging Face, so you can now use it to prepare your own documents for retrieval and to dictate your messages. Hugging Face offers both through a few of its inference providers only, which is why you get a shorter list to choose from there than you do for chatting.
+- Added a model list for Hugging Face. Until now you had to type the name of the model yourself and hope you got it right, down to its capitalization. AI Studio now loads the models your chosen inference provider actually offers, so you pick one from a list and cannot end up with a model that provider does not serve.
 - Improved the safety of plugin symbols: AI Studio now shows the symbol of a plugin in isolation, so nothing inside a symbol can reach the rest of the app.
 - Improved how much memory AI Studio needs. Working with large documents used to grow the app to several gigabytes, and on macOS that memory was never handed back. AI Studio now stays at a fraction of that and returns memory to your system. This matters most on devices with little memory, such as a Raspberry Pi.
 - Improved the preview for large documents. It now shows you the beginning of your document instead of loading all of it, so the dialog opens right away. Your complete document still goes to the AI.
@@ -15,6 +17,8 @@
 - Improved which models you get to choose from when chatting: models you cannot chat with are now hidden. This is most noticeable with a gateway such as LiteLLM, which offers you everything its providers have, including video, live speech, and audio models.
 - Changed the model list of GroqCloud. Models which cannot be used for chatting, such as the speech and the safety models, no longer show up among the chat models. The speech models now appear where they belong, in the settings for speech-to-text.
 - Changed how plugins your organization rolled out are protected. They can no longer be deleted or edited in AI Studio, which already applied to sharing and replacing them. This also covers plugins staged for a test: such a test now ends by restarting AI Studio or by removing the staged files, instead of through the plugin page.
+- Fixed the Hugging Face provider, which had stopped working. Hugging Face changed the way requests are addressed, and AI Studio still used the old way, so chatting failed with a puzzling error about the message format. Chatting works again, and you can now reach far more inference providers, among them Z.ai, Groq, Cohere, DeepInfra, and Baseten. You may also leave the choice to Hugging Face and let it pick the fastest or the cheapest provider for you, which switches to another one when your first choice is unavailable. Should a provider not offer the model you selected, AI Studio now tells you so in plain words instead of reporting a technical problem. The providers Hugging Face no longer runs are gone from the list; if you had picked one of them, AI Studio asks you to choose again.
+- Fixed the abilities shown for Google's Gemma models. AI Studio did not recognize them at all and treated every one of them as a text-only model, so images, reasoning, and tool usage stayed hidden even though Gemma 4 handles all three.
 - Fixed assistants created by the Assistant Builder being named after an internal placeholder, such as "Model decides", when you left the display name empty. The model now picks a fitting name instead.
 - Fixed AI Studio reading a document to the end even after you closed its preview. Closing the dialog now stops that work immediately.
 - Fixed AI Studio holding on to finished chats, presentation images, and plugin data. It releases them now, so memory no longer grows the longer you keep the app running.