As X targets Nitter, a developer revives Berners-Lee's Giant Global Graph argument

A 2007 essay finds a new audience

In 2007, Tim Berners-Lee published an essay titled The Giant Global Graph, musing that people were asking for their relationships with other people to rise above individual documents and websites, so that any other site or program could work with that information. Nearly two decades later, that essay is circulating again, and the trigger is a fight over access to social posts.

The revival comes from a blog post by developer Paul Frazee, titled "Select * from Internet.blogposts," which surfaced on the Hacker News front page. Frazee, who says he works on the AT Protocol, argues that the essay feels timely because X is sending cease-and-desist letters to Nitter, a lightweight frontend that let people read posts on the service without logging in.

The long closing of the platform APIs

Frazee recaps a familiar arc. In 2007, Twitter's API was famously open, and thousands of developers built clients, tools and analytics on it for free. As the network grew, developers stopped being an asset, he writes: rate limits arrived, then pricing tiers, login requirements and technical blocks on workarounds, and eventually letters from lawyers. He notes that Meta ran a similar playbook roughly a decade earlier, to the point where it is hard to remember that Facebook and Instagram ever had APIs worth building on.

Nitter's history tracks that closure. According to the post, it started on X's APIs, switched to reading public web pages when the API shut, and now faces demands that its source code be removed, with a program that merely displays public posts treated as a circumvention tool under computer-crime law. Frazee also cites Internet Archive founder Brewster Kahle, who has spent more than a decade calling for the web to be "locked open."

His conclusion is blunt: the walled-garden dynamic will not change from the inside, so the realistic option is to build fresh networks. He points to continued growth in atproto and the resilience of ActivityPub as signs that builders are already doing so.

The query the web cannot run

The post's framing device is a database metaphor. SELECT * FROM users is how a program asks a database for everything it knows about its users. The web, Frazee argues, has no equivalent. It is a few dozen companies, each holding its own data and granting access through APIs that answer a fixed menu of predefined questions, at a pace and on terms those companies choose.

He lists the resulting constraints. An API only answers questions someone already thought to expose; if your product needs posts from the people your followers follow, ranked by how often they get quoted, no endpoint exists and none is coming. Responses arrive in awkward shapes, so a two-million-follower account means 20,000 paginated round trips, ruling out anything that must feel instant. Data cannot be joined across services, and you cannot build an index over data you do not hold, which rules out search, ranking, recommendations and much moderation tooling.

A serious service, he argues, needs the whole dataset rather than a view onto it, arriving live as it changes, indexable however the product demands, writable, and guaranteed in a way that no single company's quarterly priorities can revoke. Users, meanwhile, need identity that outlives any one app and exports that remain operable, not static archives.

atproto's answer: share many databases

The AT Protocol's approach, as Frazee describes it, is not one shared database but many: a network of personal data servers that applications interact with. Rather than sending complex queries across the network, data is replicated onto event logs, and each application keeps its own local copy to query however it needs. Writes go to a user's personal data server and then replicate out to other applications, with an optimistic local update pattern for responsiveness.

He offers scale figures as of his writing: 46.1 million accounts, 24.5 billion records including 3.15 billion posts and 17.4 billion likes, 500 to 1,000 write events per second, and more than 5,000 personal data servers. A service called Jetstream lets developers subscribe to the live event stream, and Bluesky, Tangled and Leaflet are named as applications working this way. The entire network, he notes, can be read from a laptop without asking anyone's permission.

Why it matters

The legal pressure on Nitter extends platform control beyond APIs to code that merely renders public pages, which the post reads as the endgame of two decades of enclosure. The deeper argument is that the open web's survival hinges on data access, not just open formats: whether the next generation of social networks treats its corpus as a commons that any developer can query, replicate and index, or as a private asset metered at the gate. Berners-Lee's 2007 vision of relationships transcending individual sites is, in this telling, less a prediction than an unfinished design brief. It is worth noting the case for atproto is made by someone building it, the account of X's legal actions comes from a single post, and X's side of the dispute is not represented in the source.