Datasets to support agents with specific context data (vector stores, basically)

[danielrosehill]

I'm very excited to see that my watchtower seems to be finally working and I've got the latest version with the agent feature addition!

Here is the accompanying feature that I've seen deployed in a large number of agent focused tools on which would be a great add-on as you're already implementing RAG.

Knowledge Management & Data Stores

While having a single collection of files in RAG and being able to inject them one by one is a start, it's much more useful to be able to aggregate these into discrete knowledge stores and to connect one or multiple to specific agents.

I made a YouTube video yesterday demonstrating how I did this in another frontend if it's of any interest. But I'm sure the idea is pretty self-explanatory.

Here's an example of a dataset that I created for my food and drink preferences (funny data just to demonstrate the idea, I hope!)

A very useful feature in this regard is being to edit these files directly and on an ongoing basis:

Potential Use Cases & Contexts

While simple, what can be achieved with this architecture is extremely powerful (Although, obviously, we all know that the most exciting thing AI can do is provide us with beer recommendations!)

For example, in the context of a job search, I could create a batch of knowledge for my job search, like my resume, my ongoing applications, list of dream companies, etc.

Then I create a job search assistant agent connected to that data store and perhaps also connected to my general personal contacts data store.

Being able to connect to multiple different data stores, in fact, is where this becomes insanely useful (And being able to group datasets into groups would also be very helpful).