This message was deleted.
# proj-agent-canvas
d
This message was deleted.
b
I'm been experimenting with interactions and cam up with this two days ago. Ignore the colors and non themed elements. I'm working on the top area. a kind of stacked card.
Here's an example of an action above the field
the idea is to bring the user actions related to field content above the control area
f
cc @wooden-umbrella-84098
b
Here are some experiements adding more buttons and differentiation. Again, ignore the colors and theme. I'm just architecting elements to see how they appear in the feed (evaluating what appears to be helpful and clear)
w
So @breezy-kilobyte-39430 this particular message is not part of the message history--it's a high level state that isn't part of the workstream
it's something like "you ran out of credits, go refill"
✅ 1
or "you're disconnected from the internet"
b
Do you have a list of them?
w
I don't think we have a definitive list, unless there's a pattern in the translations?
f
@breezy-kilobyte-39430 some example keys of the type of messages we'd render through that banner are:
STATUS$ERROR_LLM_AUTHENTICATION
,
STATUS$ERROR_LLM_SERVICE_UNAVAILABLE
,
STATUS$ERROR_LLM_INTERNAL_SERVER_ERROR
,
STATUS$ERROR_LLM_OUT_OF_CREDITS
,
STATUS$ERROR_LLM_CONTENT_POLICY_VIOLATION
Here is an example of how we would currently render an error observation within the chat interface. An example would be if the agent tried to write/edit a file that could not be decoded for some reason.
🙌 1
These type of errors are visible to the agent and allow it to act accordingly, unlike the error messages we would show via the banner
As for the designs you shared, I personally like the clean styles. I can see ongoing actions be rendered above the field (e.g., Running command <some command>) and then render the result (observation) in the chat interface
The only thing I'm not sure about are the suggestions under the field. We do provide suggestions, but they're really just pre-defined messages to fill in the input for the user to send for now. That is why I cc'd Robert
b
Those are different experiments, specifying a specific mode of talking with the ai