Cursor launches Origin, a GitHub-competing code hosting platform

Cursor's pitch for Origin is not that GitHub is bad at hosting code - it's that GitHub was never built for the traffic pattern AI agents now generate. Origin ships as a git forge covering repos, pull requests, code browsing, and GitHub sync, explicitly framed around what Cursor calls 'agent scale' ^[1]. The company backs that framing with numbers: it claims Origin can sustain roughly 296,000 clones per hour and about 22 commits per second on a single repository ^[3], throughput well beyond what a team of humans clicking commit would ever produce. Cursor also disclosed that 35% of pull requests merged inside its own product are opened by agents running autonomously in cloud VMs ^[4], a figure meant to justify why a forge purpose-built for machine-speed traffic matters now rather than hypothetically.

The technical backbone for that bet is Graphite, the code-review startup Cursor acquired in December 2025 ^[6]. Graphite's stacked pull-request workflows and its co-founder Tomas Reimers reportedly sit behind Origin's development ^[4], which explains why Origin isn't simply a GitHub clone - it inherits a review model built around chains of small, dependent diffs rather than single monolithic PRs. That matters because agents tend to produce exactly that shape of change: many small, sequential commits rather than the batched, human-reviewed pull requests GitHub's UI was designed around. Origin doesn't ask teams to abandon GitHub to test this thesis, either - bidirectional sync means a comment made in Cursor posts to GitHub and a reaction on GitHub shows up in Cursor within seconds ^[1], letting the new host prove itself on live repositories before anyone commits to it as the system of record.

Origin didn't launch in a vacuum - it landed three days after SpaceX closed its $60 billion all-stock acquisition of Cursor, folding the company into a division called SpaceXAI ^[7]. That sequencing changes what Origin means strategically. A company that already controlled the editor where agents write code, and had ties through xAI to the model running those agents, now also controls the platform hosting the resulting code. Origin compounds that by defaulting to on for every paid user - it is opt-out, not opt-in, and only enterprise administrators can turn it off ^[3].

That default-on posture is what turns an infrastructure story into a governance one. Critics have pointed out that a single vendor now holds the editor, the host, and the model in one corporate family, concentrating leverage over a developer's entire workflow ^[3]. Public reaction has leaned skeptical of the ownership change specifically - less about whether Origin works technically, and more about a single company now holding the keys to where code is written, hosted, and potentially trained on. Whether that skepticism is justified or not, it's now baked into how Origin gets evaluated, separate from anything the product itself does.

Timing did a lot of the marketing work Cursor didn't have to pay for. Origin began rolling out on August 17, 2026, and roughly 3.5 hours later GitHub suffered a six-hour-42-minute global outage - its seventh status-page incident in just fifteen days ^[4][8]. Cursor engineer Matt Palmer addressed the overlap directly, saying the team 'were going to ship this earlier, but GitHub was down' ^[4]- an admission that the outage, if anything, delayed rather than prompted the launch, even as commentary treated the two events as one story.

Coverage seized on that framing anyway. VentureBeat characterized the coincidence as GitHub's outage 'exposing an opening' for Cursor to step into ^[5], and the juxtaposition - a credible, well-funded challenger going live just as the incumbent stumbled publicly - did more to establish Origin as a serious competitor than any feature comparison could have. It's a reminder that competitive narratives in developer tooling are often set less by roadmaps and more by which platform happens to be down on launch day.

For a product now hosting source code by default for every paid Cursor user, Origin shipped with a conspicuous gap: no published policy on data retention, data residency, subprocessors, or whether Origin-hosted code gets used for model training ^[3]. That gap is landing right as suspicion about who benefits from that code is already circulating in developer communities - some commenters have speculated openly that any future training use would primarily feed Grok, the xAI model tied to Cursor's new ownership, rather than serving Cursor's own product. That's community suspicion, not a confirmed fact - Cursor hasn't stated what the hosted code is used for, in either direction.

Cursor's adoption strategy softens that risk in the short term without resolving it. Origin treats GitHub as the ongoing source of truth for synced repositories, letting teams try the new host on live projects without migrating away from GitHub outright ^[4]. That's a low-friction wedge, not a forced choice - and it means the terms question doesn't have to be answered before developers start using it, only before they commit fully. Origin also isn't entering an empty market: mature self-hosted and hosted alternatives to GitHub already exist, which means its case has to rest on being meaningfully better for agent workflows specifically, not simply on being new.