da-z / llamazing Goto Github PK

Is there any way to limit the risk when opening up OLLAMA_ORIGNS to * given that it is basically removing any CORS protection? That would allow any computer to connect to your OLLAMA server potentially.

have you tried any variations of OLLAMA_ORIGINS = ...*

Hi,

Please enable browser local storage, currently the chat(s) are disappeared after refreshing the browser. using https://my.llamaz.ing/

Best,

I just came across llamazing and it seems very nicely done. I have been working on adapting a different ollama front-end to support my concept for RAG and am wondering if I should switch to llamazing. I am more of a back-end developer though I have done some work with React in the past. More recently, I've done some work with Svelte, which I think I like better, but I could consider switching back to React to use this code.

But before I do any of that, I am wondering if you have any thoughts for extending this project to support RAG? We could start by just defining an interface to hook into the chat request/response. Something like:

import type {
    ChatRequest,
    ChatResponse,
} from "./interfaces.js";


interface Hook
{
    onRequest(request: ChatRequest): Promise<ChatRequest>;
    onResponse(response: ChatResponse): Promise<ChatResponse>;
}

export class DefaultHook implements Hook
{
    async onRequest(request: ChatRequest): Promise<ChatRequest>
    {
        return request;
    }

    async onResponse(response: ChatResponse): Promise<ChatResponse>
    {
        return response;
    }
}

The first RAG implementation would just use the hooks to write each new user message and assistant message to the vector store.

Next we modify the Request by doing a semantic search of the vector store filter the ChatRequest messages[] to include only the top 3 most semantically relevant request/response pairs. This would make it easy to play with and see that the filtering is working. The idea is that if carry out a short conversation on one topic, then switch to digression topic, then switch back to the original topic, the digression should be omitted.

Then we would need a way to ingest documents, and change the filtering to include passages from documents.

Does this interest you?

Thanks for supporting local storage,

I would like to ask if you are interested in enable conversations history like most of LLM UIs, so we can switch between conversations or start a new one while keeping all the conversations storied in the local cache.

Best,