Blog Post

BibiGPT Team

[BibiGPT Growth Series] Episode 1 | How AI Summaries for Bilibili Were Born

Welcome to Episode 1 of the BibiGPT Growth Series!

This series documents the evolution of BibiGPT, shedding light on the behind-the-scenes decisions, technical experiments, and iterations that brought the product to life. Today we travel back to the early days, revisiting the video “BibiGPT: AI Summarizes Bilibili Videos Automatically.”

Early interface

Although BibiGPT has since undergone major redesigns, the ideas and lessons from that first release still matter. Let’s rewind to where it all began.

The Birth of BiliGPT: AI Summaries for Bilibili

When GPT started making headlines, I (Jimmy) felt compelled to build something tangible. Bilibili was overflowing with high-quality content, but extracting key insights wasn’t easy. That sparked the idea: What if AI could automatically summarize Bilibili videos?

The result was BiliGPT (Bilibili GPT), powered by GPT-3, created to help people grasp video highlights quickly.

BiliGPT early interface

Two Ways to Summarize

In the original demo we showcased two workflows:

  1. Copy & Paste the URL
    Paste any Bilibili link into BiliGPT’s input box, click “Summarize,” and wait a few seconds. The tool grabbed subtitles (if available) and generated a concise summary.

    Paste URL

    In the video we summarized a clip about using ChatGPT as the “ultimate coworker,” capturing points on reading, writing, note-taking, coding, and more.

  2. URL Hack
    Change bilibili.com in the address bar to jimmylv.cn while watching a video.

    URL hack

    Hit enter, and the page redirected to BiliGPT, automatically launching a summary job. We demoed it on a video about “productivity addiction,” and it worked like magic—highlighting concepts like play vs. tool modes, protecting attention, and automation.

    Summary results

Managing API Keys and Costs

Calling OpenAI APIs incurs real costs, especially as usage grows. Back then, my $18 free credit evaporated quickly. To make BiliGPT sustainable, we added an option for users to provide their own OpenAI API key.

User API key field

Keys were stored locally in the browser—never on the server—giving heavy users full control while easing cost pressure on the project.

Mobile Convenience: iOS Shortcuts & Web Hacks

Many people watch Bilibili on their phones, so we built mobile-friendly solutions:

  • iOS Shortcut – Install it from the BiliGPT site (b.jimmylv.cn), then use the Bilibili share sheet to run “Summarize Bilibili Video.” The shortcut sent the link straight to BiliGPT—no app switching required.

    iOS shortcut flow

  • Android Web Trick – Change m.bilibili.com to jimmylv.cn in the browser address bar to trigger the same flow. Simple and effective.

Early Roadmap & Vision

Whiteboard sketch

While shipping the demo, we already had a wishlist:

  • Support more platforms (Zhihu, Douban) and content types.
  • Explore deeper GPT-3 applications, including prompt engineering and private data integrations.
  • Empower more people to harness AI in daily workflows.

Even then, the vision was clear: make AI-powered knowledge extraction accessible to everyone.

Looking Back

That first release captured the excitement of tapping into GPT and applying it to a real-world pain point. BiliGPT has since matured into today’s BibiGPT with multi-platform support, advanced models, and robust integrations—but the mission is unchanged: help people learn faster, with less friction.

Thanks for joining us on this flashback. Up next in the Growth Series:

[BibiGPT Growth Series] Episode 2 | YouTube Support, Smart Timestamps, and Notion Login

Related episodes:

Original video: https://www.bilibili.com/video/BV1fX4y1Q7Ux/