Genie does not support chat with `converstation_id` for continuous converstation

[adyah]

Issue
We are trying to migrate to databricks-ai-bridge and we are using Genie extensively.
What we observed is that ask starts new conversation every time.

Possible Fix
Provide support for converstation_id

[adyah]

I am interested in providing a PR with the fix.

Thanks

[sunishsheth2009]

Hey @adyah , thank you for reporting this issue.
We are curious to understand the usecase here. converstation_id is used in Genie so we don't need to sent the entire input_history back to genie.
The reason we sent the entire history is because customers could be using multiple agents in a multi-agent setup. Now if we just sent conversation_id to the genie, it might miss all the intermediate steps that could have happened in between calls to genie. That is why always sending the input_history would make sure that genie gets the entire context.
Curious to hear your usecase and thoughts.
cc @smurching

[adyah]

Hi @sunishsheth2009
Thanks for responding.

I want to make use of

    @mlflow.trace()
    def ask_question(self, question):
        resp = self.start_conversation(question)
        return self.poll_for_result(resp["conversation_id"], resp["message_id"])

If I want to make continued conversation. i.e I already have conversation_id how to achieve it.

For now we have this flow with direct api calls.

# pseudocode

if no converstation_id
   start_conversation
if conversation_id
   create_conversation_message
   return execute_sql (when status == Complete)

Please guide us, as we are looking to use databricks-ai-bridge for genie.
Above is our requirement in pseudocode
Thanks

[sunishsheth2009]

Thank you for sharing your usecase. We are wondering if it would be easy to directly use our databricks-sdk for this usecase:
https://databricks-sdk-py.readthedocs.io/en/latest/workspace/dashboards/genie.html
We also internally plan to migrate to this for databricks-ai-bridge. Let me know if that works for you since that always more customization and is future proof.

[adyah]

Here is our final goal.

We want to use https://github.com/databricks/databricks-ai-bridge/blob/329a064452c6c59c568de2edfbd952aa6de67481/integrations/langchain/src/databricks_langchain/genie.py#L40C5-L40C15

GenieAgent and rest of the things already being handled.

This is repo is doing everything what.
We need flexibility on two places

1. Converstation_id injection
2. Dataframe return should also support row = Array(array), column=Array(string). For now it just renders markdown table.

If these both are solved then I think it would be great and will add extensibility to this project.
But I will convey your msg to my team

Thanks

[smurching]

@adyah thanks, did you end up getting to try the Databricks SDK approach? Curious if you and the team found any gaps there - happy to follow up on getting those addressed if so