Extract Data tool I'm trying to better understand how the Extract Data tool works. In the doc and when asking @help, it's seems like Extract Data is able to ensure exhaustivity (like Include Data) while not being limited by the context window (like Search)... it seems impossible to me but I don't find more precise information about it. Without knowing this I have a hard time trusting the outputs of this tool (even if I see that it's indeed way more exhaustive than Search... but still, is it 90% exhaustive or 100% ? I'm not sure...). Is it because we just fill the context with the json objects resulting from the data extraction that it takes less space than including everything from the source ? In this case, we still do not guarantee exhaustivity, just that "more data is processed compared to Search" and we can't trust the output when it says that for example "68% of tickets says XYZ". It is still 68% of tickets processed, but not 68% of all existing tickets if there are a lot of them. Thanks !
Hi Ghyslain 👋 The extract data tool will let the agent create a table with a bunch of column with info it is looking for. It will loop trough all of the documents in the datasource attached and store the values for each doc in a csv table. Once it finished looping through the data, it will use the csv (and only the csv) to answer. If you want you can download the csv it generated by clicking on "inspect tool" and the file at the bottom. It is 100% exhaustive with the only limitation being that in total it can process ~500k token of documents at the moment. Hope it helps clarify!
