Description
Collectively, these tasks support improved output rendering for model responses in the Model, Prompt, and/or Flow runner(s) and the Trace viewer.
One way to prioritize these is to start with the model (and flow) runners, because once that is completed, any "model step" in a trace can be viewed in its rendered state by opening that step in the model runner. For example, run a flow, open the trace, find the model step and click "open in model runner".
One common theme that is emerging is the idea of "multiple" ways to view a given model response. For example, a single text message returned from the model could be rendered:
- In plain text
- As rendered markdown
- As separate streamed "chunks"
Thus far we have preferred the "rendered" view (for example, images, tool requests and responses). We should continue to do so, however for text specifically there is value in offering a plain text option. Particularly for streamed responses, additional thought should be given to the UX (i.e. consult with Marissa).
Sub-issues
Metadata
Metadata
Assignees
Type
Projects
Status