archivr

things on the internet don't stay. posts get deleted, videos go private, sites shut down, and links you saved years ago quietly stop working. archivr keeps a full copy instead (the page, the video, the thread, the file) on your own machine, so it's still there when the original isn't.

work in progress, and the one i'm most proud of.

The archivr web app: a list of saved pages, videos, a playlist, a PDF and notes, with a summary of a Turkish vs Azerbaijani video in the side panel — pages, videos, a playlist, a PDF and notes in one list. the summary on the right was written from the video's subtitles.

what it saves

videos: YouTube videos, shorts, playlists and whole channels, with subtitles. pick the quality, or keep only the audio. run a playlist again later and it downloads just the new videos.
web pages: a complete copy of the page as a single HTML file, images included, saved through a real browser. ads and cookie banners can be stripped, and reader mode keeps only the article.
X posts and threads: the text, the photos and videos, and how the replies connect to each other.
Instagram, TikTok, Reddit and Facebook posts.
files and notes: upload any file, or write a note in plain text or Markdown without a link at all.

A saved copy of Maciej Cegłowski's talk The Website Obesity Crisis, opened inside archivr — a saved page opens from your archive, not from the original site.

A 3Blue1Brown playlist expanded in archivr, showing each video saved inside it — a saved page opens from your archive, not from the original site.

finding things, and keeping them

search covers titles, links, tags and summaries.
tags can nest, like languages/turkic or web/archiving.
collections group entries, and decide who else on your server can see them.
the same file saved twice takes up space once.

there's no cloud service in between. your archive is a folder of ordinary files plus one SQLite database, and you can export any of it to HTML or Markdown. if archivr stopped existing tomorrow, you would still have everything you saved.

summaries

press Generate on an entry and archivr writes a short summary: one line on top, a paragraph under it, and a few tags. it reads pages, notes, X threads, and YouTube videos through their subtitles, so you can tell what a 25-minute video is about without watching it.

summaries and their tags go into search, so a video turns up for what it's about, even when its title says nothing.

nothing is summarized unless you ask. it can be written by the Anthropic or OpenAI API, any OpenAI-compatible server (one on your own computer works too), or the claude and codex command-line tools you already have.

rightbelow: archivr's summary of the Wikipedia article on link rot.

archivr's side panel showing a generated summary and tags for the Wikipedia article on link rot

browser extension

right-click a page, an X post, a video, or some selected text, and pick Archive. the same capture dialog opens on top of the page, with the quality picker and all the options, and the capture runs on your server while you keep browsing.

it connects to your server with a single API token, and it only sends your cookies for a site when you tick "use my cookies". Chrome only for now, and it's not in the Web Store yet, so you load it unpacked. (GitHub)

The archivr capture dialog opened by the browser extension on top of a Langfocus YouTube video about the Caucasian languages — saving a Langfocus video straight from YouTube.

MCP server

archivr-mcp lets Claude Code, Claude Desktop, or any other MCP client use your archive: search it, read summaries, save links and files, tag and organize. it talks to your server with an API token, and it can run read-only. (GitHub)

what's in my archive about link rot? one or two lines per entry.

archivr · search_entries "link rot"
archivr · search_entries "dead links"
archivr · get_summary × 2

Your archive has two entries about link rot:

Archiving URLs · Gwern.net (web page, archived 2026-10-03): Gwern Branwen's system for archiving every link he cites, since link rot steadily destroys references and public archives miss many pages.
Link rot - Wikipedia (web page, archived 2026-08-17): Defines link rot and summarizes studies on how often links break, including dead references in court opinions and Wikipedia.

save https://jeffhuang.com/designed_to_last/ and tag it like my other archiving stuff.

archivr · list_tags
archivr · capture_url "https://jeffhuang.com/designed_to_last/"
archivr · tag_entry "/web/archiving"

Saved https://jeffhuang.com/designed_to_last/ and tagged it /web/archiving. I picked that tag because it's where your other archiving captures live: it already has 5 entries, and it sits under /web.

a real session with Claude Code, shortened.

running it yourself

with Docker Compose:

git clone https://github.com/thegeneralist01/archivr.git cd archivr mkdir config && cp docker/config.example.toml config/archivr-server.toml docker compose run --rm archivr archivr init \ /data/archives/main /data/archives/main/.archivr/store --name "Main Archive" docker compose up -d

then open http://localhost:8080 and create your account. saving X posts needs a cookies file, and docker-compose.yml shows where it goes.

with Nix:

nix run github:thegeneralist01/archivr#archivr -- init ./my-archive --name "My Archive" nix run github:thegeneralist01/archivr#archivr-server

the server reads archivr-server.toml from the current directory, which lists your archives (the README has an example). there's a NixOS module too. the full setup guide is on GitHub.

‹ back to projects

`←` `a` `h`	previous tab
`→` `d` `l`	next tab
`?`	this help
`esc`	close