sbot
Simple Archiver Bot -- a suckless web archiver written in C.
sbot creates self-contained archives of web pages and entire websites.
Every resource -- CSS, images, fonts, scripts -- is fetched and inlined
directly into the HTML as base64 data URIs. The result is a single file
(or directory of files) that renders perfectly offline, with no external
dependencies, forever.
WHY
Web pages disappear. Link rot is real. The average web page has a
half-life of about two years. Bookmarks break, articles vanish,
references evaporate.
sbot solves this by creating archives that are:
- Self-contained. Everything is inlined. No external requests needed.
- Human-readable. Output is standard HTML. Open it in any browser.
- Permanent. No database, no server, no special viewer. Just files.
- Metadata-rich. GWTAR headers record provenance, date, and source.
MODES
SINGLE PAGE ARCHIVE
sbot https://example.com/article
Archives a single page in GWTAR format (Gwern Web Tar Archive). This
is the default mode and the most common use case. The output is one
.gwtar.html file containing:
- A GWTAR metadata header (HTML comment) with title, source URL,
domain, author, archive date, and generator version
- The full HTML with all CSS stylesheets inlined as style blocks
- All images, fonts, and media encoded as data: URIs
- A completely self-contained document that renders identically to
the original
GWTAR format is ideal for:
- Archiving individual articles, blog posts, and essays
- Preserving references and citations
- Building a personal web archive / digital library
- Saving pages before they disappear behind paywalls or get deleted
WHOLE SITE ARCHIVE
sbot -r https://example.com
Recursively crawls an entire website and archives every page. The
output is a directory tree that mirrors the site structure, with each
page saved as a self-contained HTML file. Internal links are rewritten
to relative paths so navigation works offline.
Features:
- BFS crawl order. Breadth-first traversal ensures important pages
(closer to root) are archived first.
- Same-domain only. Never follows links to external sites.
- robots.txt compliance. Respects Disallow rules and Crawl-delay
directives by default. Override with -R.
- Depth control. Set maximum crawl depth with -d to limit scope.
- Rate limiting. Configurable delay between requests to be polite to
servers (default: 1 second).
- Progress reporting. Periodic status lines showing pages archived,
queue depth, and elapsed time.
- Graceful degradation. Failed resources are skipped; the crawl
continues.
This mode is ideal for:
- Archiving entire blogs or documentation sites
- Creating offline mirrors of reference material
- Preserving small-to-medium websites wholesale
- Building browseable offline copies of sites you depend on
USAGE
usage: sbot [-vrR] [-d depth] [-o dir] [-a author] url
-v verbose output
-r recursive (crawl entire site)
-R ignore robots.txt
-d depth max crawl depth (default: 5)
-o dir output directory
-a author site author name
EXAMPLES
Archive a single article:
sbot https://example.com/blog/post
Archive with author metadata:
sbot -a "John Doe" https://example.com/article
Crawl a blog, max depth 3:
sbot -r -d 3 https://blog.example.com
Verbose crawl to custom directory:
sbot -v -r -o ./my-archive https://docs.example.com
Crawl ignoring robots.txt restrictions:
sbot -r -R https://example.com
GWTAR FORMAT
Every archived page includes a GWTAR (Gwern Web Tar Archive) metadata
header as an HTML comment at the top of the file:
<!--
================================================================
GWTAR ARCHIVE
================================================================
Title: Example Article
Source URL: https://example.com/article
Domain: example.com
Author: John Doe
Archived by: Kris Yotam
Archived on: krisyotam.com
Archive date: 2026-02-14
Generator: sbot/0.3.0
Format: GWTAR (Gwern Web Tar Archive)
================================================================
-->
This header provides full provenance tracking: what was archived,
where it came from, who archived it, and when.
RESOURCE INLINING
sbot inlines all resources to create truly self-contained archives:
Resource Type Inlining Method
------------------ ------------------------------------------------
CSS stylesheets Fetched and inserted as style blocks
Images Base64-encoded as data:image/* URIs
Fonts Base64-encoded as data:font/* URIs
Other media Base64-encoded with appropriate MIME type
Resources that fail to fetch are silently skipped -- the archive
degrades gracefully rather than failing entirely.
BUILD
Requires libcurl development headers.
Arch Linux:
sudo pacman -S curl
Debian/Ubuntu:
sudo apt install libcurl4-openssl-dev
Build:
make
Install to /usr/local/bin:
sudo make install
Clean:
make clean
CONFIGURATION
All configuration is compile-time via config.h:
Setting Default Description
----------------- ------------ -----------------------------------
USER_AGENT sbot/0.3 HTTP User-Agent string
CONNECT_TIMEOUT 30s Connection timeout
REQUEST_TIMEOUT 60s Total request timeout
MAX_REDIRECTS 10 Maximum HTTP redirects to follow
MAX_DEPTH 5 Default recursive crawl depth
RATE_LIMIT_MS 1000ms Delay between requests
MAX_FILE_SIZE 50 MB Maximum size per resource
OUTPUT_EXT .gwtar.html File extension for archives
Edit config.h and rebuild to change any setting. This is the suckless
way -- no runtime configuration files, no environment variables, no
hidden defaults.
ARCHITECTURE
archiver.c Main entry, page archiving, CSS inlining, link rewriting
crawl.c URL queue (BFS), visited set, URL normalization
fetch.c HTTP fetching via libcurl
parse.c HTML parsing, resource extraction, image inlining
robots.c robots.txt fetching, parsing, rule matching
util.c Memory wrappers, string ops, base64, MIME types
config.h Compile-time constants
Single external dependency: libcurl. No XML parsers, no HTML5 parsers,
no JavaScript engines. The HTML parsing is deliberately simple --
regex-based extraction of src, href, and url() references. This handles
the vast majority of real-world pages and keeps the codebase small and
auditable.
PHILOSOPHY
sbot follows the suckless philosophy:
- Written in C99 with POSIX.1-2008
- Minimal dependencies (libcurl only)
- Configuration through config.h (edit and recompile)
- Small, readable codebase
- Does one thing well
LICENSE
MIT/X Consortium License. See LICENSE for details.