🦙 Ollamate: SPIKE Data Storage

[Gary-H9]

The purpose of this Spike is to understand how best to store conversations generated in Ollamate.

The current thinking is that this should be stored in AWS RDS. This is based on a number of assumptions.

Call and responses are outlined in here.

This spike does force some architectural consideration as to how we send/recieve/show entire conversations to the user.

A chat request, with history is of this form :

curl http://localhost:11434/api/chat -d '{
  "model": "llama3",
  "messages": [
    {
      "role": "user",
      "content": "why is the sky blue?"
    },
    {
      "role": "assistant",
      "content": "due to rayleigh scattering."
    },
    {
      "role": "user",
      "content": "how is that different than mie scattering?"
    }
  ]
}'

The question therefore is if we ought to store the entire JSON in messages, associate this with a something like a conversation_id and update this in the database each time a new call / response is made.

It would be advisable to research the approach of other applications such as openwebui to understand their approach.