A PopClip extension for ChatGPT

[nick]

You could enable extension debug output by pasting the following command in Terminal:

defaults write com.pilotmoon.popclip EnableExtensionDebug -bool YES

then Quit and restart PopClip.

(note, substitute com.pilotmoon.popclip-setapp if using Setapp edition)

Then open Console app and set the following filters process: PopClip, category: Extension

Now set Console to “Start” & “Now” you should see the API calls going in and the responses. You can also add debug using print(), e.g. such as print("varname:", varname); in the code if that helps.

[rdouma]

Thanks @nick, that helped. Turns out my paid ChatGPT account didn’t cover API calls; I needed to add that separately. Now it seems to work.

[nick]

Ahh! Good to know. I have the paid ChatGPT too and wasn’t sure whether that linked up with the API service. So they are separate.

Looks like the API comes with 3 months free credit then you have to pay on a per usage model at $0.02 per 1000 tokens (1 token = about half a word).

[rdouma]

Interesting! I’m still getting used to ChatGPT and to think of new possibilities, it’s such an amazing tool. It would be nice to have a more versatile plugin where you can predefine multiple prompts, like “Summarize” or “Translate to Dutch”, essentially any prompt that someone can come up with. Just floating there idea out here…

[Tweaker]

This is great. Thanks for putting the extension together @nick! I’ve been using the shortcut that’s been floating around but super handy to have the direct code (davinci-003) in popclip.

[Tweaker]

@nick whelp it appears openai literally just released an api to chatgpt which is big news - time to update the script!

[nick]

Ooh. Here are the docs. A quick first attempt using the simplest possible approach:

// #popclip
// name: GPT-3.5
// icon: iconify:logos:openai-icon
// language: javascript
// after: paste-result
// entitlements: [network]
// options: [{
//   identifier: apikey, label: API Key, type: string,
//   description: 'Obtain API key from https://platform.openai.com/account/api-keys'
// }] 
const openai = require("axios").create({
  baseURL: 'https://api.openai.com/v1/',
  headers: { Authorization: `Bearer ${popclip.options.apikey}` }
});
const params = {
  model: 'gpt-3.5-turbo',
  messages: [{ "role": "user", "content": popclip.input.text }]
};
const { data }  = await openai.post('chat/completions', params);
return popclip.input.text + data.choices[0].message.content;

It uses the newly released gpt-3.5-turbo model, but it only works as a “one shot” question i.e. it doesn’t remember what the previous chat was. I need to look a bit more closely to see how to do that.

[nick]

OK, here is a snippet for the “real thing” ChatGPT, that you can actually chat with.

Update, 3 Mar 2023: A pre-packaged version of this extension is now available.

To use it, you’ll need to sign up for an OpenAI platform account, and get your own API key from https://platform.openai.com/account/api-keys .

// #popclip extension for ChatGPT
// name: ChatGPT
// icon: iconify:simple-icons:openai
// language: javascript
// module: true
// entitlements: [network]
// options: [{
//   identifier: apikey, label: API Key, type: string,
//   description: 'Obtain API key from https://platform.openai.com/account/api-keys'
// }]
const messages = []; // history of previous messages
async function chat (input, options) {
  const openai = require("axios").create({
    baseURL: "https://api.openai.com/v1",
    headers: { Authorization: `Bearer ${options.apikey}` },
  });
  messages.push({ "role": "user", "content": input.text });
  const { data } = await openai.post("/chat/completions", {
    model: "gpt-3.5-turbo", messages
  });
  messages.push(data.choices[0].message);
  return input.text.trimEnd() + "\n\n" + messages.at(-1).content.trim();
};
exports.actions = [{
  title: "ChatGPT: Chat",
  after: "paste-result",
  code: chat,
}, {
  title: "ChatGPT: Reset",
  icon: "iconify:game-icons:broom",
  requirements: [],
  after: "show-status",
  code: () => messages.length = 0 // clear the message history
}];

(The above block is an extension snippet – select the text then click “Install Extension” in PopClip.)

How to install it:

Note there are two actions, Chat and Reset.

Chat (main icon): Type your message in any text field, select it so PopClip appears and choose the Chat action. After ChatGPT responds, type your response and repeat.

Reset (broom icon): click it whenever you need to start a new chat and forget the previous responses.

I had to break out the module syntax (which I still haven’t properly documented) for this one because the extension itself has to remember the past messages in the chat and provide the whole thing to the API each time. But it’s really a very straightforward API to use.

[Mikemaan]

@nick

// #popclip extension for ChatGPT
...

Brilliant!

[nick]

Since the ChatGPT extension seems to be working well and is getting pretty popular, I’ve created an “official” downloadable version and added it to the main extension directory.

The new version includes a few refinements over the one in this thread. See forum post.

[jberrebi]

Bizarrely, it doesn’t work for me.

I put an API Key and I’m getting no answer. I tried in Notes, Word…etc…

See here:

Ps: I have a ChatGPT Plus account but not a Paid API account but I understand I don’t need one?

[rdouma]

Fairly sure you need to have a paid API account. For me it also only started working once I had configured that.

[jberrebi]

What do you say @nick ?

[nick]

@rdouma is correct, and it is the most likely explanation. Have you checked the paid credit status of your OpenAI account @jberrebi? Failing that, there are debug instructions in this post.

[jberrebi]

I added my credit card to my OpenAI account and it works indeed

This extension is really a killer app.

Thanks @nick and @rdouma

[uudricc]

I need a proxy to access OpenAI. Is it possible to use axios proxy?

The proxy config is not working properly:

const openai = require("axios").create({
    baseURL: "https://api.openai.com/v1",
    headers: { Authorization: `Bearer ${options.apikey}` },
    proxy: {
        protocol: 'http',
        host: 'localhost',
        port: 7890,
    }
  });

I see that some people have found a workaround using HttpsProxyAgent, but this package seems not applicable in Popclip extensions.

[nick]

It’s a little in-depth for me to dive into right now, but if Axios supports proxy config like that, I would expect it to work. However, the log shows that the underlying XMLHttpRequest is going directly to the API, so it isn’t using the proxy.

Modules loaded within popclip extensions are bundled inside PopClip’s executable. The current bundle axios version is 0.26.1. There are only a small number of other modules bundled, but none for networking.

However, I notice your proxy uses HTTP, and unfortunately, PopClip will refuse to make an outgoing connection that isn’t https anyway, even to localhost. So even if you get the proxy set-up working, it won’t connect

[jberrebi]

I got an access to ChatGPT4 API.

I updated the code and installed the new extension.

I tried to use it but I’m getting axios errors

Error: Request failed with status code 429
axios.js.lzfse:667:24
axios.js.lzfse:923:23
axios.js.lzfse:67:13

Result: error: Error: Request failed with status code 429 (axios.js.lzfse:667:24)

‘paste-result’ was specified but there is no result text

according to OpenAI, they sent back an answer but I never got it…

Any idea?