PHION
Files 10 Uptime 309d Latest 0d

Social media and news without the algorithm (Techno-fascism, Part 4)

By burt Published 2026-07-27 Class Power & Society Read 12 min

Part 4 of a five-part series on techno-fascism and the tools that push back against it. Part 1 covered the machinery of surveillance capitalism, Part 2 covered private messaging, and Part 3 covered browsing and search. This part covers where you get your social feed, your video and your news: the Fediverse, RSS, and privacy front-ends for platforms you haven’t left yet.

On the evening of 18 January 2025, TikTok switched off its own servers in the United States. A law requiring ByteDance to sell the app or shut it down had taken effect at midnight. Apple and Google pulled it from their app stores; Oracle and Akamai, which hosted TikTok’s US infrastructure, dropped it. Anyone who opened the app was told, in a full-screen message, that the new law had made the app unavailable for now. By TikTok’s own count, that message reached over 170 million American accounts. It stayed up for roughly 14 hours, until president-elect Trump signed an executive order pausing enforcement for 75 days and shielding TikTok’s service providers from liability. The app came back. But for most of a day, 170 million people’s video feed, DMs and creator income depended entirely on a decision made by one company and one government, with no fallback and no export path that mattered in the moment.

That’s the extreme case, but the underlying condition is universal to mainstream social platforms: a single company decides what you see, collects everything about how you look at it, and can revoke access to both at will. The algorithmic feed and the centralised control point are the same piece of infrastructure. This piece is about the alternative infrastructure — federated social networks, RSS, and privacy front-ends for platforms you’re not ready to leave — and what each one specifically changes.

What the algorithm optimises for

Part 1 covered the ad-auction side of surveillance capitalism. The feed side runs on a related but distinct mechanism: engagement-optimised ranking. TikTok’s own research describes its recommender as operationalising multidimensional signals — completion rate, dwell time, interaction frequency — to predict what keeps a given account watching. A 2024 CHI study using donated TikTok interaction data found the recommender responds quickly to selective engagement: accounts that liked a handful of drinking-related videos were served progressively more of them, with measurable downstream association to self-reported drinking and vaping behaviour. None of this requires the platform to know your name. It requires a continuous, granular feed of what you watched, for how long, and what you did next — collected specifically so the ranking model can be tuned on you personally.

Fediverse software, RSS and front-ends don’t fix engagement-optimised ranking as a concept — nothing stops a Mastodon client or an RSS reader from adding one. What they remove is the precondition: a single company sitting between you and the content, positioned to collect that feed of behavioural data by default and required to for the product to work.

The fediverse

“Fediverse” refers to a family of independently run servers that speak a common protocol, ActivityPub, a W3C recommendation since 2018. Mastodon (microblogging), Pixelfed (photos), PeerTube (video) and Loops (short video) are all separate pieces of software built on it — different apps, one protocol, and accounts on any of them can follow and reply to accounts on any other.

How a post travels

Each ActivityPub account lives on a server — an “instance” — and has an inbox and an outbox, the same shape as an email address. Posting is a Create activity dropped in your outbox; your instance then delivers a copy directly to the inbox of every server that has a follower of yours, addressed with the same kind of headers used for email delivery, minus a central switchboard. Mastodon’s own documentation describes the model as functioning much like email does at the macro level. There’s no equivalent of TikTok’s or Instagram’s central ranking server that every post and every read passes through, because there’s no single server the whole network runs on.

On a closed platform, every post and every read passes through one company’s servers before delivery — the same servers that build your engagement profile. On the fediverse, your instance delivers a post directly to the inbox of each server with a follower, with no central copy in the middle.

One protocol, several apps

Software What it’s for Scale (2026)
Mastodon Microblogging (Twitter/X-shaped) ~10.5 million registered accounts, roughly 750,000–1 million monthly active, across thousands of instances
Pixelfed Photo and short-video sharing (Instagram-shaped) Roughly 100,000 monthly active users across 680–750 servers as of mid-2026
PeerTube Long-form video hosting (YouTube-shaped) Over 400,000 videos and roughly 60,000 users across hundreds of instances, built by the French non-profit Framasoft
Loops Short-form video (Vine/TikTok-shaped) Donation-funded (Patreon, Ko-fi, a 2025 Kickstarter); explicitly rules out venture funding and advertising

Mastodon’s own numbers illustrate how tightly this network tracks the mainstream platforms it’s an alternative to. It ended 2021 with 294,000 monthly active users and 2,527 servers. After Elon Musk completed the Twitter acquisition in October 2022, Mastodon logged its largest single-day signup surge to that point — 70,000 accounts in one day — and closed out 2022 at 1.8 million monthly actives and 9,686 servers, with donations up 488% year on year. Various trackers put the peak slightly higher, around 2.5–2.6 million monthly actives in November 2022, a level the network hasn’t returned to since; by 2026 it had settled back to the 750,000–1 million range. The pattern is consistent with what it claims to be: a release valve people move to when a closed platform does something alarming, not a platform most people choose from a standing start.

Moderation without a company

There’s no fediverse-wide terms of service, because there’s no single company to write one. Each instance sets its own rules and moderates its own users; instances federate, limit or fully block (“defederate” from) each other individually. This is how the network handled Gab’s 2019 attempt to join as an ActivityPub server — most of the network blocked it outright — and it’s the same mechanism at work in the ongoing, informal Fedipact, where a large number of instance admins have pledged to defederate from Meta’s Threads over concerns about Meta’s data practices and moderation standards.

Threads has been shipping ActivityPub interoperability since March 2024 and expanded it through 2025 to include a dedicated fediverse feed and cross-network search. On 8 January 2025, the same week Meta announced it was replacing third-party fact-checking with a crowdsourced Community Notes model, Mastodon CEO Eugen Rochko criticised the change publicly and warned Mastodon users who cross-post from Threads that content violating Mastodon’s stricter rules would still draw enforcement on the Mastodon side. That exchange is the model working as intended: no company can unilaterally set the moderation policy that applies to instances it doesn’t run, but a company that dwarfs the rest of the network in size can still change what shows up in your federated timeline by changing its own rules. Scale asymmetry doesn’t disappear just because the protocol is open.

Owning your account

Mastodon supports account migration between instances without losing your social graph. It requires setting your new account as an alias of the old one, which triggers a Move activity: every follower whose instance implements it is automatically unsubscribed from the old account and subscribed to the new one. Mastodon’s documentation is specific about the limits — your posts stay on the old server and don’t move, your own follows/blocks/mutes need a manual CSV export and import, and there’s a 30-day cooldown before you can migrate again. It’s not a full backup-and-restore. It is a mechanism, absent from every closed platform discussed above, for taking your audience with you when you leave a server you no longer trust — the exact capability TikTok’s 170 million US users didn’t have on 18 January 2025.

Where “decentralised” gets complicated

Bluesky markets itself as decentralised and is often grouped with the fediverse, but it runs a different protocol (AT Protocol, not ActivityPub) with a different architecture. Its own designers describe it as three layers: a Personal Data Server (PDS) that stores your repository of posts and can in principle be self-hosted; a Relay that crawls every known PDS and republishes the combined stream as a “firehose”; and an AppView that consumes the firehose to build the actual timelines, search and moderation the app shows you. In theory anyone can run any layer. In practice almost nobody runs a PDS, and running a Relay is the real barrier: Christine Lemmer-Webber, co-author of the ActivityPub specification, documented that a full Relay needed about 1 terabyte of storage in July 2024 and roughly 5 terabytes by November 2024 — upwards of US$500 a month even on cheap block storage, and closer to US$55,000 a year on conventional cloud storage pricing. Lemmer-Webber’s conclusion is that “a large corporation or two always have to sit at the center of Bluesky,” with credible exit — the ability to leave and take your identity with you — as the more honest claim than decentralisation itself. That’s a real, useful property, and distinct from what ActivityPub instances offer.

RSS

RSS (and its sibling format Atom) predates the algorithmic feed by a decade and works on the opposite principle: a feed is a plain XML file at a fixed URL, your reader polls it on a schedule, and you see every item, in the order published, from every source you subscribed to — no ranking model, no engagement signal sent back to anyone, because there’s no account to send it to.

Its near-death is a cautionary tale about the same centralisation problem this whole piece is about, just a decade earlier. Google Reader was, for most of the RSS-reading public, the RSS reader. On 13 March 2013, Google announced it would retire the service on 1 July 2013, citing declining usage despite a loyal following. Alternatives existed, but Reader’s shutdown removed the default on-ramp at the moment algorithmic social feeds were becoming the obvious substitute, and RSS usage never recovered its earlier mainstream footing. The lesson generalises: a free, single-company-run reader is exactly as durable as that company’s continued interest in running it.

The current self-hosted options don’t have that failure mode, because you run them yourself. Miniflux is a minimal, fast Go application with essentially no configuration surface; FreshRSS is more feature-rich, with themes, extensions and multi-user support. Both are free and open source, both speak the Fever API that most third-party mobile RSS apps use for sync, and both can be self-hosted with Docker in a few minutes:

docker run -d --name miniflux \
  -p 8080:8080 \
  -e DATABASE_URL="postgres://miniflux:secret@db/miniflux?sslmode=disable" \
  -e RUN_MIGRATIONS=1 \
  -e CREATE_ADMIN=1 \
  -e ADMIN_USERNAME=admin \
  -e ADMIN_PASSWORD=changeme \
  miniflux/miniflux:latest

If self-hosting isn’t appealing, managed options like PikaPods run Miniflux from roughly US$1/month. The one caveat worth knowing: RSS itself carries no tracking by default, but a feed can still embed a tracking pixel or truncate content to force a click-through to the publisher’s site — the format’s privacy properties depend on what the publisher puts in the feed, not just on how you read it.

Front-ends for closed platforms

Not everything worth watching or reading has a fediverse or RSS-native version. YouTube and Reddit don’t, and neither did Twitter/X before it locked down further. Privacy front-ends are open-source web interfaces that sit between you and the original platform: they fetch the content server-side and serve you a page with no ads, no tracking scripts and no login tying the view back to your identity, without you ever loading a script from the origin platform in your own browser.

Invidious and Piped both do this for YouTube — proxying video, hiding your IP from Google, and (in Piped’s case) adding SponsorBlock integration to skip sponsored segments automatically. Both are locked in a permanent cat-and-mouse fight with YouTube, which actively IP-blocks known frontend instances; by 2026 that pressure had collapsed Invidious’s public instance list from dozens down to a handful, while Piped’s federated instance model and faster patch cycle left it with more surviving public instances. Redlib plays the same role for Reddit — a Rust rewrite of the abandoned Libreddit, using OAuth credential spoofing to emulate Reddit’s own official client and dodge its rate limits, proxying all media through its own server so nothing loads from reddit.com directly.

Nitter, the equivalent for Twitter/X, is the reason to take this category seriously rather than treat it as a permanent fix. It worked by pulling tweets through Twitter’s anonymous “guest account” API, unauthenticated by design. On 26 January 2024, Twitter/X removed guest account access outright. Nitter’s maintainer announced the project dead the following month; the flood of scraping traffic from clients still trying to reach shut-down instances briefly resembled a denial-of-service attack against what remained of the network. The surviving forks require logging in with a real X account to keep functioning — which reintroduces exactly the account-ban risk the front-end existed to avoid. A front-end only works for as long as the origin platform’s API surface allows it to, and that surface is entirely the origin platform’s decision.

Building a practical setup

  1. Pick a Mastodon instance from joinmastodon.org/servers, ideally a small, topic- or region-specific one rather than the largest general instance — smaller instances mean one admin with less scale to abuse, and an easier migration later if needed.
  2. Or self-host one. A small single-to-medium-user Mastodon instance runs comfortably on a US$20–30/month VPS (4GB RAM minimum; image/video uploads are the main memory spike), or via a managed host like masto.host from US$6/month.
  3. Add PeerTube or Pixelfed accounts on existing instances for video and photos, rather than running your own — federation means you don’t need to run every layer yourself.
  4. Self-host Miniflux or FreshRSS for anything that still publishes a feed — most blogs and many news sites do, even ones that don’t advertise it (check /feed or view-source for a <link rel="alternate" type="application/rss+xml"> tag).
  5. Run Piped or Invidious as personal infrastructure, not a public instance — self-hosted via Docker, or pointed at one reliable public instance for read-only use, treating any public instance as liable to break without notice.

Limitations

None of this replicates a closed platform’s reach. The fediverse’s total active user count is still a rounding error next to Instagram or TikTok, so posting there reaches a much smaller, more technical audience — a real cost for anyone whose reason for being on social media is being seen. Content moderation quality varies enormously by instance and by admin availability; a one-person instance can vanish or go unmoderated with no recourse. Mastodon’s private mentions are not end-to-end encrypted — they’re visible to the admins of both the sender’s and recipient’s instances, unlike Signal or Matrix’s encrypted rooms covered in Part 2, so they shouldn’t be treated as a substitute for actual private messaging. RSS only works for sources that still publish feeds, and a growing number of platforms — Instagram and X included — publish none. Front-ends live entirely at the mercy of the platform they front for, as Nitter’s death demonstrated, and any of Invidious, Piped or Redlib could face the same fate on a similar timeline. Self-hosting anything in this piece — an instance, a reader, a front-end — makes you your own single point of failure: if you don’t apply updates, your own server is the new expiring dependency.

Sources