Replies: 3 comments 3 replies
-
Not yet -- the simple approach, which is to continue adding user messages to the end of the conversation does not work well in practice because of the types of things that users can potentially ask. They may ask to filter/pivot the table, request changes to the chart, ask for some LLM-based action like extract topics. We have seen that users also somewhat expect that the LLM can "see" the chart (which requires a vision-enabled LLM). Up until now we were trying to be LLM-agnostic where you could plug in any LLM and it would function roughly similarly. Your request is not uncommon so we will likely be moving to a model where there will be "Basic" functionality that will be possible across any LLM and "Advanced" functionality like what you describe which will only be usable with the top-of-the-line LLMs. |
Beta Was this translation helpful? Give feedback.
-
In fact, making the web interface like a chat conversation would be fine, otherwise, a new page has to be opened for every question asked |
Beta Was this translation helpful? Give feedback.
-
Is there some way to use Langchain, for example, with Vanna specifically for this? |
Beta Was this translation helpful? Give feedback.
-
Does Vanna support multiple rounds of dialogue?
Ask again based on the answer to the previous question
Beta Was this translation helpful? Give feedback.
All reactions