Agents with databases

[blythed]

An agent seems to be a language model or sequential model using text, images, files and otherwise, which has a memory of
the past discussions/ actions, and can generate new output on the basis of this.

Since superduperdb has great support for flexible file-types and for language models, it would make a lot of sense to support this in some way. A simple langchain integration won't work, since the interfaces are so different to what we have.

My suggestion is that we employ the concept of "auto-regressive" model, with the additional possibility to persist output
in the database. We should allow the user full transparent control over prompts, Encoders etc.