Working in IT, I have dozens of saved Assistant conversations that go into details of tool use, bug solving, fixes, best practices etc; which required the Assistant to already dig through hundreds of web pages. In many cases that's already great, but when bigger picture assistance is required, I'd like to point the Assistant at my own conversations, so that it can make use of detailed conversations that have already happened.
I would consider it enough to just enter one or more tags in the prompt in order to trigger the inclusion of matching conversations. To judge matching converstations, I would use an embedding of the conversations and/or their parts, then match the embedding of the prompt to conversations and include them for the LLM to use, in addition to any web search results, if web search is on.
For instance, let's say we're talking about command line operations to work with PDFs documents, and I have a dozen conversations on different operations, and they are all tagged as "#PDF"; then I ask to generate a script for a complex workflow including the tag "#PDF" and the Assistant would match and include those existing conversations and maybe include anything from the web of use.
