About

This website holds an archive of metadata for over 60 million Twitch videos, and video content for over 630,000. While this project was designed initially for speedruns, a large variety of popular streams were saved, as well as the metadata for every public video on the site (as of April 23, 2025; ongoing archives to begin at a later™ date).

Usage

On the Search page you can input a link to a video, a link to a streamer, or a video title search term. You can optionally filter to only certain types of videos or only videos with archives available. Press Enter or click the Search button to submit your search.

Once you submit your search, you'll see a list of results. You can click on each one to bring up more information, including where to watch the broadcast (if available).

Wayback Machine archives

Internet Archive currently does not include CORS headers on its Wayback Machine endpoints which makes it impossible to playback in an embed. Additionally, the video streams consist of hundreds or thousands of video segments, so it's impossible to provide a direct download link.

As a workaround for easy playback, I've added an option to toggle on a CORS HTTP proxy, although these services can be slow or block certain kinds of requests. If you're facing issues, you may be able to install a CORS patcher addon for Firefox or Chrome but please be aware that CORS helps protect your user data from being exfiltrated to other services, so I would recommend only using an addon when necessary.

You can also download the video stream file and load it in VLC. MPV sorta works but doesn't buffer any video segments so it's a poor experience. If you need to download the full video stream, yt-dlp works.

Backstory

On February 19, 2025, the livestreaming platform Twitch announced that effective immediately, streamers could no longer permanently save more than 100 hours worth of content to their account. Beginning on April 19, later postponed to May 19, they would retroactively enforce this rule by automatically deleting content with less-viewed content, leaving only the 100 most-viewed hours of content on every channel.

Within days of the announcement, and having remembered the aftermath of lost speedruns following the mass privatization of unlisted YouTube videos in 2021, I (lexikiq) began scraping all videos on speedrun.com and cross-referencing them against Twitch's API to find which videos appeared to be at risk of deletion.

Simultaneously, arkiver of the ArchiveTeam began working to identify all the GraphQL queries used by the Twitch app to fetch the metadata for videos and began scraping all 2bn+ IDs in bulk using AT's distributed warrior system.

While this scrape was in progress, I finished my list of speedruns as well as a larger list of highly viewed videos on the site as a whole using a scrape of my own, and shared them with arkiver who ultimately was very generously able to offer nearly half a petabyte of storage on the Internet Archive's servers to save video content. In the end we were able to save over 630,000 videos and the metadata of tens of millions more. Due to technical issues, some archived data is currently unavailable through this tool. While I am slowly recovering it, you can also manually recover data using the instructions shared by arkiver on the AT wiki here.

A few speedrun mirrors are also available on YouTube, notably some longer videos that failed in the AT Warrior's queue. This collection is much more limited however, totaling only ~3700 videos, and many of them facing issues from copyrighted audio.