Chat History Management

[giusedroid]

Issue #35 , if available:

Description of changes:

Front-end, Back-end and prompt changes to enable users chat history management.
This only makes use of Local Storage.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[kirtandudhatra]

Overall very useful change!!

My suggestion:

• We are saving the chat history in local storage and passing it to the lambda with <history> </history>. Instead of that, we can prepare the list like this on the frontend side:

[
        {
          role: "user",
          content: [{ text: "first question" }],
        },
        {
          role: "agent",
          content: [{ text: "first answer" }],
        },
        {
          role: "user",
          content: [{ text: "second question" }],
        },
        {
          role: "agent",
          content: [{ text: "second answer" }],
        },
];

And pass this array to the lambda.

• Now in lambda, instead of compiling the prompt with history, we can simply add the array received by frontend into the message we are passing in converse API. For example,

    const conversation = [
        {
          role: "system", <- BY defining this, LLM will understand that this is the ultimate task we want to achieve
          content: [{ text: "prompt without any mention about history" }],
        },
    ];
    conversation = conversation.push(...history)

This approach will take advantage of the existing ability of the converse API instead of doing that work on our side.

[giusedroid]

ah interesting! I'll try to make these changes asap :D

[giusedroid]

bumping into this error: An error occurred while invoking the selected model. 1 validation error detected: Value 'system' at 'messages.1.member.role' failed to satisfy constraint: Member must satisfy enum value set: [user, assistant]

Having a look. Looks like system is not supproted 🤔 maybe I need to upgrade the bedrock sdk or something?

[giusedroid]

Implemented 😸

[kirtandudhatra]

https://docs.aws.amazon.com/bedrock/latest/userguide/conversation-inference.html#conversation-inference-examples

converse api supports system prompt as one of the argument. Neat!