We have been using this plugin internally for meetings, which has been great for gathering feedback and evolving it based on real use cases. In the last two weeks, the highlights are:
AI Noise Suppression
Noise suppression was completely re-done, with several changes. We now offer controls to handle the standard WebRTC features:
- Noise Suppression
- Auto Gain Control
- Echo Cancellation
Furthermore, the plugin now ships three different AI models for those who need better noise suppression, from low to high respectively:
- RNNoise
- DTLN
- DeepFilter3
Direct calls
Users can now initiate 1-to-1 or 1-to-many direct calls via a button on the user card. This uses ephemeral rooms infrastructure we built, which will also be used to handle integrations in the future, like with the events plugin.
Invites
Right-clicking a room allows you to generate invites, and it uses your conversation history in that room to suggest users.
Sound packs
Resenha sounds now follow your preferred sound pack from New chat notification sound packs.
Room UX
Right-click to copy link to room, widget tiles auto-sizing and collapse, widget positioning above the site header, shift+click room support, accuracy and better performance on the voice activity indicator.
Android Camera Switching
You can now switch between your Android phone front and back cameras seamlessly mid-call.
Mesh connectivity warning
The first time a user tries to connect to a room that uses WebRTC mesh, we present a warning that explains potential IP exposure.
And last but not least, two experimental features 
Live Subtitles
We added opt-in live subtitles, powered by state-of-the-art AI models running on device.
When enabled, we download a 2.5GB AI model that is run using WebGPU, so compatibility is restricted to fast and modern devices.
There is a lot to polish and fine-tune on this feature, so let me know how it goes.
Transcription
Users with live subtitles enabled can now start transcriptions from those, which are backed to a new topic draft. After finishing the recording, you get a composer with all the saved transcriptions using the same format chat quotes use.