Get In, We're Leaving GitHub

Why people are suddenly serious about leaving GitHub

Theo starts from a place of real frustration: GitHub isn’t just annoying anymore, it feels unsafe to trust with your code. He points to reverted merges and outages lasting days, then sets the bar for alternatives: solid Git hosting, PR workflows, community features, CI/CD, and ideally stability, self-hosting, and some AI-native future path.

GitLab: the obvious answer that falls apart on contact

GitLab is the big-name alternative, but Theo says using it feels like talking to someone who bikes to work: technically valid, practically miserable. He walks through painful UX examples — back navigation that hides repos behind loading layers, a project page where the README is buried 75% down, release pages with “88% complete” labels that explain nothing, and commit history that becomes an infinite-scroll punishment ritual.

The real problem with GitLab is deeper than ugly screens

He then zooms out: GitLab is huge, old, and hard to evolve — 528,000 commits and a codebase he clocks at roughly 12.7 million lines, including 3.88 million Ruby lines and 1.16 million JavaScript, apparently still on Vue 2. That leads to his larger thesis: GitLab isn’t a new generation of product, it’s just another Gen 2 GitHub-like system, the way Azure is to AWS — familiar, enterprise-friendly, but not meaningfully better.

Bitbucket disqualifies itself by telling you exactly what it is

Theo barely has to stretch here because Atlassian’s own pitch does the work: “Git solutions for teams using Jira.” He mocks the pricing comparison built around squeezing $15 per engineer per month and says the whole value proposition is basically “already paying Atlassian? here’s more Atlassian,” with uptime and bundled support being the only genuinely useful differentiators he can find.

Forgejo and Codeberg unexpectedly become the stars of the video

This is where the video flips. Theo goes in expecting another ugly open-source compromise, but comes away impressed: Forgejo’s codebase is much smaller, mostly Go, modern, and clearly maintainable, while Codeberg’s hosted version feels plain but functional in the ways that matter. The release page wins him over completely because it shows the date, commit count, and actual changes up front — “ugly as sin,” in his words, but exactly what a developer needs.

He likes it enough to put money behind it

Theo keeps testing for failure points — issues, code review, themes, image embeds, Actions compatibility — and keeps finding something passable or surprisingly solid. The governance story seals it: Forgejo forked after Gitea went more commercial, sits under Codeberg e.V., and is run as democratic nonprofit free software, which prompts Theo to donate $1,200 on the spot and commit to $400 per month.

The Gen 3 bet: agent-native infrastructure, not another GitHub clone

With the current-generation pick basically settled as Forgejo/Codeberg, Theo turns to what’s next. He’s most excited by Pierre’s code.sto, which he frames as infrastructure built for agent-scale throughput: 9 million repos in 30 days, peaks of 15,000 repos per minute for three hours, all while GitHub struggles under AI-driven repo and PR volume. His demo contrast is classic Theo — GitHub CLI takes nine questions and multiple loading pauses to create a repo, while code.sto does it in one line.

Graphite, Entire, and the sad part nobody can fix

He sees Graphite as a company that learned the hard way how limiting GitHub’s APIs are, then began pulling core functionality onto its own infrastructure before being acquired by Cursor. Entire, founded by former GitHub CEO Thomas Dohmke and backed by a huge $60 million seed, is exploring how agent context and “why” should live alongside code, while Zed’s DeltaDB is similarly questioning whether Git snapshots are enough. But Theo ends on the emotional truth: even if the tooling gets better, the one shared community graph where every package, maintainer, issue, and contribution lived on GitHub is breaking apart, and no aggregator can fully replace that loss.