nbonamy/witsy

AI Summary: Debug a bug in the Witsy AI assistant where the Deep Research mode incorrectly prompts the user to download a PDF file instead of directly reading its content. The issue seems to be related to how the application handles `Content-Disposition: attachment` headers and may involve modifying how PDFs are processed within the Deep Research functionality.

AI Summary: The task is to fix a bug in the Witsy AI assistant where a Vega chart creation plugin is not properly registered, preventing it from generating charts. This involves adding the plugin to the application's plugin registry to make it accessible to the user interface and the large language models (LLMs).

AI Summary: Integrate support for the Granite-speech-3.3-8b and -2b models into Witsy, allowing users to download and utilize these models for local speech-to-text transcription. This will involve creating a similar implementation to the existing Whisper model support, handling potential VRAM requirements for the larger model, and possibly exploring API options if they become available.

AI Summary: Implement support for Soniox Speech-to-Text (STT) in the Witsy application. This involves adding Soniox as an option in the settings and the speech-to-text screen, allowing users to utilize both asynchronous and real-time transcription capabilities via the Soniox API. The implementation should handle API key integration and potentially offer a choice between asynchronous file transcription and real-time streaming.