Is it possible for the Assistant API to retrive files based on user’s context?
For example - I have 3 files in the vector store connected to the Assistant AI.
File 1 - Topic A
File 2 - Topic B
File 3 - Topic C
I want the assitant to reterieve the relevant files only when usrr mentions about that particular topic. If user mentions about Topic A and B, then only File 1 and 2 should be retrived.
If yes, is this possible through playground /instructions or need to use API for this?
It does not retrieve “files”. There is no more file, only text chunks.
AI writes a search query; top match chunks previously extracted from docs are returned based on AI similarity.
AI then can answer better.
In that case , is it correct to say once the files are added to the vector store, then they are created into text chunks (embeddings) and merged into a single hypothetical file on which Similarity works on
There is not really even a single “file” exposed. Everything in a vector store is intermixed, and you cannot request answering from one document.
The only semblance of document is that the chunk data will return a source filename along with text contents, and that you can still manage backend with removal by filename.
Understood, Thanks for the help!