Is there any way for passing document IDs in recall function?

[ielgezabal]

Hi! I'm trying to pre-filter documents in recall function but I can't find the way. Is there any way to pass document ids to recall and reflect functions? Thanks!!

[nicoloboschi]

Hi! I'm trying to pre-filter documents in recall function but I can't find the way. Is there any way to pass document ids to recall and reflect functions? Thanks!!

Hi! There's no way because the graph retrieval wouldn't make much sense. What is the use case you want to build? Likely you can use multiple bank IDs for it or share some of them (here's an example https://hindsight.vectorize.io/cookbook/recipes/support-agent-shared-knowledge )

[ielgezabal]

Hi! Thanks for the answer. I have an application similar to NotebookLM, organized by notebooks with various sources in each. Currently, each notebook has a ‘bank_id’, and each source is stored as a document. The issue is that if I need to select only specific sources within a notebook to perform an action, I currently have no option to do so. The recall() function operates on the entire notebook, which can lead to a loss of efficacy and efficiency depending on the user’s prompt. Should I create a separate bank_id for each source?”

[nicoloboschi]

you could both, at cost of double retain - which might not be a problem if your dataset is not that big.

If you need to work on elaborating multiple notebooks of different kind to get summaries, pattern detection or just really reviews, you need a single bank id.

If you need to find things specific to a notebook, yes, you need a bank_id per notebook.

If you need both, you can call retain in the per-notebook bank AND in the shared bank

we are open to new suggestions and we are actively working into improving the architecture for these use cases.

[nicoloboschi]

hey @ielgezabal we just added document tagging in 0.3.0, you might want to check it out. you can tag a document (and its memories) and the pass it to filter in only memories matching that tag

[ielgezabal]

Hi!!! I was just reading the changelog. It looks so interesting, i will try it