Candid — AI video summaries with timestamps & chat for YouTube™

Gemini watches the video, not just captions: AI summaries with timestamps beside the YouTube™ player. Free with your own Gemini key.

Candid — AI video summaries with timestamps & chat for YouTube™ Candid puts an AI summary beside the video you are watching: a short overview, then the key points grouped under headings, each with a timestamp that jumps the player to that moment. Google's Gemini watches the video itself, so it hears what is said and sees what is on screen: slides, charts, price cards, tables. It does not need the video to have captions. What you need • A free Gemini API key from Google AI Studio. The extension shows you where to get one and checks it the moment you paste it. • Google's free tier allows a limited number of requests a day. Google offers the Gemini API in many countries but not all, and requires its users to be 18 or older. What you get • The full transcript, searchable, with a note under each line of what was on screen as it was said. • Ask about the video, as in a chat. Answers come from what was already read, link the moments they came from, and say plainly when the video doesn't cover the question. A new look at the video happens only when you click, and its cost is shown first. • Short, Medium or Detailed summaries, or your own instructions. • Summaries in 12 languages: English, Español, Português, Français, Deutsch, Italiano, Polski, Русский, Türkçe, 日本語, 한국어 and 中文, or in the video's own language. • Translate the transcript on your device, with Chrome's built-in translator. • Summarize when you ask, or automatically for every video. • Optionally, write summaries with your own Cloudflare, Hugging Face or Anthropic key. Private by design • No account, no server, no analytics. Your keys and your saved summaries stay in your browser. • Video links and transcripts go only to the AI provider you use, directly and under your own key. Honest about what it read • A table in a summary says where its numbers came from, read on screen or said, and counts how many were really read off the screen. A claim that doesn't hold up is marked down. • The panel names the models that read the video and wrote the summary, and how long each took. • Before you start, it shows how long a summary should take, from measured runs, "give or take" a stated margin. Where nothing was measured, it shows no number at all. • A summary you come back to is shown as saved, with anything that has changed since, and is never passed off as new. YouTube and Gemini are trademarks of Google LLC. This extension is not made or endorsed by Google. Source code, privacy policy and contact: https://github.com/sevadelorean12/candid