@mightymeld/codemirror-copilot

codemirror-copilot

This CodeMirror extension lets you use GPT to autocomplete code in CodeMirror. It let's you call your API with current code (prefix and suffix from current cursor position) and let your API return the code to autocomplete.

Demo

https://copilot.asadmemon.com

Installation

npm install codemirror-copilot --save

Usage

import CodeMirror from "@uiw/react-codemirror";
import { javascript } from "@codemirror/lang-javascript";
import { inlineCopilot } from "codemirror-copilot";

function CodeEditor() {
  return (
    <CodeMirror
      value=""
      height="300px"
      extensions={[
        javascript({ jsx: true }),

        // Implement a function that returns a promise that resolves to the prediction
        inlineCopilot(async (prefix, suffix) => {
          const res = await fetch("/api/autocomplete", {
            method: "POST",
            headers: {
              "Content-Type": "application/json",
            },
            body: JSON.stringify({ prefix, suffix, language: "javascript" }),
          });

          const { prediction } = await res.json();
          return prediction;
        }),
      ]}
    />
  );
}

You also need to implement an API that returns the prediction. For above code, here is an example API in Next.js that uses OpenAI 's gpt-3.5-turbo-1106 model:

import OpenAI from "openai";

const openai = new OpenAI({
  apiKey: process.env.OPENAI_API_KEY,
});

async function completion(
  prefix,
  suffix,
  model = "gpt-3.5-turbo-1106",
  language,
) {
  const chatCompletion = await openai.chat.completions.create({
    messages: [
      {
        role: "system",
        content: `You are a ${
          language ? language + " " : ""
        }programmer that replaces <FILL_ME> part with the right code. Only output the code that replaces <FILL_ME> part. Do not add any explanation or markdown.`,
      },
      { role: "user", content: `${prefix}<FILL_ME>${suffix}` },
    ],
    model,
  });

  return chatCompletion.choices[0].message.content;
}

export default async function handler(req, res) {
  const { prefix, suffix, model, language } = req.body;
  const prediction = await completion(prefix, suffix, model, language);
  console.log(prediction);
  res.status(200).json({ prediction });
}

API

inlineCopilot(apiCallingFn: (prefix: string, suffix: string) => Promise<string>, delay: number = 1000)

Provides extension for CodeMirror that renders the hints UI + Tab completion based on the prediction returned by the API.

The delay parameter is the time in milliseconds to wait before calling the apiCallingFn after the user stops typing. Default value is 1000.

The extension also implements local caching of predictions to avoid unnecessary API calls.

clearLocalCache()

Clears the local cache of predictions.

Local Development

In one terminal, build the library itself by running:

cd packages/codemirror-copilot
npm install
npm run dev

In another terminal, run the demo website:

cd website
npm install
npm run dev

Acknowledgements

This code is based on codemirror-extension-inline-suggestion by Shan Aminzadeh.

License

MIT © Asad Memon