A work in progress. The design is being drafted right now. Watch the repository and come shape the project! We'll use the Issues section as a mailing list.
I'm looking for developers to join me!
Here is a quick presentation on the project.
Produce a permanent, future-proof, faithful archive of all the web pages you visit.
- Archived pages should be as static as possible, but they should include dynamic content as loaded in a browser. À la archive.today. An archival copy of everything exactly as you saw it, plus some (that you didn't see -- expanded comments, etc.).
- There should be broad support for plug-ins, both JS injected in pages and for the backend.
- Storage should be permanent, deduplicated and indexed.
- Full-text search on entire archive, with OS-specific UI integration.
- Browse your history even in offline mode!
Here are just some ideas on how to build it. Have a better idea? Share it!
- A local browser extension should store the URLs
- The browser-loading part should probably happen in PhantomJS or Selenium
- Then JS plugins should be injected.
- Then CSS should be embedded, forms neutralized, JS removed, links edited (?), images fetched and DOM snapshotted.
- Then the result should be stored encrypted and deduplicated Tarsnap style.
Suggestion: The local browser extension should handle not only storing the URLs, but also initiating and orchestrating the archiving events. This enables two things: avoid double-downloading; and storing the content exactly as visited by the user.
Suggestion: Use Selenium (not PhantomJS). Because: Tighter browser integration. This enables both store-exactly-as-visited, and easier integration of ar.chiv.io plugins (even with dynamic switching plugins on/off from the browser interface).
Suggestion: [Use the MAFF file format] (http://maf.mozdev.org/maff-file-format.html) for archival. It has/includes several useful properties, such as:
- compressed (ZIP)
- internal links: "Enabling off-line navigation between pages in different archives"
- Metadata:
- original URL the page was saved from; date/time of the save operation
- "arbitrary extended metadata", such as "browser's scroll position in the page, text zoom level, and more"
The [MAF browser extension] (http://maf.mozdev.org/) already implements the grunt work.
Suggestion: Store session-internal versioned-history. (This is a conceptual parallel to the versioned history that applies to a page whose content changes as it is visited repeatedly over weeks/months.)
Why? A conflict arises from the desire to save both:
- The page exactly as the user saw it
- The full content of the page, not just parts (scroll-loaded content, etc.)
This is compounded by the desire to avoid double-downloading pages (which wastes bandwidth, and alerts servers to something fishy).
Both these issues can be solved via a versioned-history approach, storing successive versions of the page as it is visited. The first layer is the page as the user sees it upon page load, and subsequent content changes can be saved as diff layers. Upon page close, a Selenium event can expand any remaining content (comments, video downloads, etc. -- as per installed plugins).
Thus, the execution happens like so:
- User loads page
- Browser extension triggers a MAFF snapshot (executing any before/after plugin-events via Selenium, if applicable)
- User interacts with page
- Browser extension detects changes and makes further MAFF DIFF-snapshots when necessary
- [Repeat steps 3-4 indefinitely...]
- User closes page
- Browser extension makes a final snapshot, in background (executing onPageClose-plugins, if applicable)
TODO: Define an heuristic for when to snapshot (how much of a change counts).
- A set of browser-specific extensions (Firefox, Chrome, ...)
- ar.chiv.io core daemon, handling deduplication, storage, etc.
- ar.chiv.io CLI tools, to manually interact with store when desired
- purge entries by site, export/import/partition stores (per browser, date, etc.)
- perform on-demand web crawls/whole-site-fetching
- A set of index-service integration plugins, exposing the data for indexing by specialized services (Tracker, YaCy, Spotlight, Windows indexer...)
- Plugins (for JS-injection etc.), interoperable between browser extension and web-crawl CLI tool
(On second thought, is the daemon even needed?)
- ~/.archivio/
- config/
- JS-plugins & their settings
- settings for index-service integration plugins
- data/
- metadata/
- hashes of each MAFF & MAFFDIFF file
- hashes of each file inside each MAFF/DIFF file
- encryption details, if used
- content/
- tld.domain.subdomain/
- pagetitle_datetime.maff
- metadata: url, UA, datetime, referrer-url, ...
- pagetitle_datetime.maffdiff
- metadata: same as MAFF
- metadata: hash/id for substrate MAFF or MAFFDIFF file (that this maffdiff increments)
- pagetitle_datetime.maff
- tld.domain.subdomain/
- metadata/
- config/
- Hooks and config for automatic snapshotting
- Plugins for JS-injection, etc.
- Play nice with other browser extensions (AdblockPlus, Greasemonkey...)
Fetch pages manually (from CLI/GUI), whether:
- single page
- whole site (follow links, stay in-domain)
- crawl (follow links, do not stay in-domain)
...with various custom settings defined by config and plugins.
- Obey Private Browsing Mode -- do not archive/index.
- Normal Firefox "Forget this site" feature → should trigger deletion from archivio store?
- Consider NOT encrypting local storage: provides a false sense of security (unless user wishes to enter password every time they access their archive). Of course, any off-site storage/backup must be encrypted.
- How to handle multiple user profiles/Chrome sign-ins? (Encrypt each with different key, and decrypt the key using the user's password...? How to coordinate user A across FF and Chrome, vs user B across both?)
Full-text search on your browsing history. A user-wide store that integrates browsing from multiple sources (browsers); but that records the origin (Firefox- and Chrome-visited pages are not conflated).
Integrate with desktop indexing services.
- MetaTracker
- YaCy
- Others...?
Both MetaTracker and YaCy support indexing HTML files. Would indexing Readability.js output provide any better results?
What backend to use for the deduplication?
How to calculate diffs on MAFF files? (Just diff each internal file?)
Work out a format for storing the MAFF DIFF files.
- Auto scroll to load more content
- Expand comments (Reddit, ...)
- Dismiss pop-ups (Quora, ...)
- youtube-dl for supported sites
- Blacklist sites - regexes
- Fetch whole site (Readthedocs, ...)
- Repo cloning (GitHub, ...)
- Run Readability and store/index the result