David decided: Added Research Mode toggle to Kerra workspace, switching model to GPT-5.4

David described the feature while viewing a screenshot of the workspace:

i want us to, in the input text bar, include a "Research Mode" toggle that, when the user turns it on, switches the model in the top banner to GPT-5.4 (even if theyre on the free plan, it shoudl show the GPT-5.4 option)

When the agent raised questions about streaming parity with Gemini, David questioned the assumption:

but why is streaming needed? go investigate how the interactive workspace ai chat works when the selected model is gemini 3 flash. does it use streaming? dont we simulate streaming by streaming the final output in some nice transition? why dont we just stick to what we already do with that?

After the investigation confirmed real streaming would be needed for GPT-5.4 parity, David approved building a proper OpenAI streaming path mirroring the existing Gemini implementation.