15. April 2024

Added Anthropic Support & Minor Changes

Claude Models are now supported with Vision & Tooling. I might need to rename the toolkit now!

After a break on working on the ChatGPT Toolkit, I sat down and added some new features and bug fixes over the weekend.

Here are the caveats.

Claude Model Support

I added all Claude models from Anthropic, with vision & tools support. You can also add a baseUrl to Anthropic if you use a third-party provider for the API.

Claude 3 Chat

Add messages to startChat

You can now specify multiple messages in advance and also take screenshots and insert them into the chat before starting it.

Here is an example:

// Name: OCRTeX
// Group: ChatGPT

import "@johnlindquist/kit";
import { startChat } from "../kenvs/chatGpt/lib/common";
import {
    createScreenshotMessage,
    userMessage,
    imageFileMessage
} from "../kenvs/chatGpt/lib/helper";

const screenshotMessage = await createScreenshotMessage();

await startChat({
    model: "gpt-4-vision-preview",
    temperature: 1.0,
    maxTokens: 4096,
    runPromptImmediately: true,
    messages: [
	    screenshotMessage, 
	    imageFileMessage("path/to/imageFile"),
	    userMessage(`Describe the images.`),
	]
});

Bug Fixes / Minor changes

• I updated the list of models by fetching them from the OpenAI API. Once a day the models are updated, so you no longer won't miss any new ones
• I fixed some rendering issues when using images, this involved storing the images in a different way, which should also be more performant for fetching the old chats

I might add Mistral Models soon.. 👀