Plans to add Gemini?

[satoshinotdead]

Bard can search the web for free! 🤟

https://github.com/ra83205/google-bard-api

[merefield]

Yes, I'd love to add support and for any emerging primary publicly-accessible LLM from Google.

Not whether but when I get around to it.

PR welcome in meantime!

[merefield]

Apologies, on review, I'm removing this from roadmap as given the general lack of community PRs I do not have capacity to entertain it.

[dillfrescott]

Awwww I think this would have been awesome!

[merefield]

I would accept a PR @dillfrescott if it's not a settings nightmare. That's why it's marked as PR Welcome and not Closed. If someone is willing to fund me I would consider doing it myself. Renaming the issue.

[dillfrescott]

Hey Robert thanks for the response. Is there a rough estimate of how much someone would have to pay for you to integrate it?

[dillfrescott]

Found this. Do you think that could work instead of having to integrate it into the plugin?

[dillfrescott]

Update: It works fantastically! updating the chatbot.prompt.post value to just %{raw} works great for Gemini!

EDIT: See updated prompt below for better performance!

[merefield]

That's nuts @dillfrescott!! ❤️. Great find!

And there's me thinking that proxies were just there to add Auth headers 😂

I wonder if there's a Claude equivalent?! : 🤔

[merefield]

Update: It works fantastically! updating the chatbot.prompt.post value to just %{raw} works great for Gemini!

Does Gemini struggle with usernames being included? Can you elaborate on what effect that has? Doesn't excluding them take away the ability of the bot to @mention?

[dillfrescott]

Hmm indeed was having issues with mentions. I hadn't tested that.

I changed the prompt to:

%{username} said:

%{raw}

You are a helpful assistant. Please respond to what %{username} said appropriately.

And now it works much better!

[dillfrescott]

I wonder if there's a Claude equivalent?

Honestly probably 😅

[dillfrescott]

As you can see it also handles topic mention requests quite well!

[merefield]

Thinking about this some more, @dillfrescott, how is this arrangement going to handle Embeddings? That's critical to getting the full power out of Discourse Chatbot via RAG.

Looks like it is on their roadmap ... but not yet implemented ... and for now I suspect this is only going to work in Chatbot Basic mode?

[dillfrescott]

I don't think it can do embeddings yet, at least with this setup. I am using a different proxy though than the one I linked, currently.

[dillfrescott]

I would totally use openai's API but gemini API is totally free to use and openai's is not.

[dillfrescott]

I think this project can do embeddings from a quick glance at it.

[merefield]

I would totally use openai's API but gemini API is totally free to use and openai's is not.

I understand, though remember to manage your spend, especially by other forum members, you can use the inbuilt quota system.

[merefield]

I think this project can do embeddings from a quick glance at it.

that's a 404 for me

[dillfrescott]

my bad i linked a fork i deleted after testing it!

https://github.com/mth93/langchain-gemini-pro-proxy

[merefield]

I really like your original suggestion though as it has been package nicely with cloud deployment buttons. Let me know if you PR embeddings and function support ;)

[merefield]

I've added the original example and the approach to the Readme