I’d like the model to view data that it can use for responses
This data would be like real-time market data and news articles
Thanks!
Hi there and welcome to the Forum!
Can you clarify what you mean by “without sending a prompt”?
Use function calling. Only when it is needed that the function to fetch real-time market data or news articles will be invoked.
It is described in a function calling documentation. You define a function to get the data in an assistant configuration. Then when you request a Run on a thread with users request it returns with the status of requires_action and information about required function and arguments that your app can process and supply required info to continue the thread run.
The other suggestions are good and helpful but I am not sure if you are aware that in order for the language model to process anything a message needs to be send. And the market data then is the prompt. Additionally I would assume that you are also adding a system prompt to tell the model what to do with the data plus a history of previous market data.
If you want to process tick data in real time a message needs to be send for each tick of data.
You cannot send a message without a prompt by design of the models.
I know that Nvidia offers some vision models for streaming data but you need to check if that matches your use case.
1 Like