Extract and Format ChatGPT Chat History as JSON #JavaScript #MDX

Below is a specific example of JavaScript code. It creates an array of objects in title and content format.

[Shown with subscription]
    .map(a => a?.content.parts).flat().filter(a => a).map(a => a.match(/### (.*)\n\n([\s\S]+)/m)).filter(a => a).map(a => { return { title: a[1], content: a[2] }})

We’ve extracted it cleanly.

(Bonus) Let’s also convert this to MDX format. While we’re at it, let’s save it to files.

// Assuming the above result is stored in variable obj
// the summary is the first paragraph of the content
obj.map(a => `---\ntitle: '${a.title}'\ndate: ${new Date().toISOString().split('T')[0]}\nlastmod: ${new Date().toISOString().split('T')[0]}\ntags: [drawing tools, AI-assisted]\ndraft: false\nsummary: '${a.content.match(/(.+)\n\n/)[1]}'\nimages: []\n---\n\n${a.content}`).forEach((a, i) => {
    // save as a file, on browser
    const blob = new Blob([a], { type: 'text/plain' });
    const url = URL.createObjectURL(blob);
    const aTag = document.createElement('a');
    aTag.href = url;
    aTag.download = `chatgpt-${i}.mdx`;
    aTag.click();
    URL.revokeObjectURL(url);
});

(The above code has not been tested, so it may not work in some cases)

That’s it - a method to extract ChatGPT conversation history as JSON and format it into any shape using JavaScript and Chrome Developer Tools.

Continue Processing in Node.js (Added)

I happened to write a quick Node.js script while writing this article, so I’ll share it.

(async () => {

  const objRaw = fs.readFileSync(`${process.cwd()}/obj.json`, 'utf8');
  const obj = JSON.parse(objRaw);

  const mdx_strings = obj.map((item) => {
    return `---
title: ${item.title}
date: ${new Date().toISOString().split('T')[0]}
lastmod: ${new Date().toISOString().split('T')[0]}
tags: [drawing tools, AI-assisted]
draft: false
summary: ${item.content.match(/(.+)\n\n/)[1]}
images: []
---

${item.content}`
  }
  ).join("\n\n\n\n");

  // split by \n\n\n\n+
  const mdx_array = mdx_strings.split(/\n\n\n\n+/);
//   ...omitted

The tedious part of creating site articles with MDX is having to think about and set filenames (URL slugs).

So recently, I’ve been automating this with the DeepL API.

Please also refer to a recent article I wrote about DeepL: Creating Glossaries with DeepL API to Specify Translation Results #Node.js

Node.js program:

• Using “deepl-node”: “^1.13.0”
• Node.js version is v22.5.1, but it should work fine with v16.6.0 or later