Analysis
Anysphere launched Origin in beta on August 17, an AI-native code hosting platform built directly into Cursor, positioning it as a direct challenger to GitHub in the layer of developer infrastructure that sits underneath every AI coding tool on the market, according to VentureBeat. Repositories are accessible through a new Codebase tab inside the editor, with URLs formatted as cursor.com/codebase/name, and the beta ships with bidirectional sync to existing GitHub repos so teams can migrate gradually rather than all at once.
The timing produced a coincidence Anysphere did not engineer but will not mind being associated with: Origin shipped Monday morning, and GitHub suffered a roughly seven-hour outage about three and a half hours later. ThinkNext and BigGo both noted the sequencing explicitly -- Origin did not cause the opening, but it landed in front of a developer audience that spent that afternoon unable to push code, exactly the pain point a hosting alternative needs to make its case.
Built for a different traffic pattern
Origin's stated performance numbers -- 296,000 clones per hour and 22 commits per second per repository -- are not framed as generic hosting benchmarks. They are framed around the traffic pattern autonomous coding agents generate: high-frequency, small commits, clones spun up by ephemeral cloud sandboxes rather than a developer's laptop. Cursor has previously cited an internal figure, reported by RuntimeWire, that 35% of pull requests merged inside the editor are opened by agents running autonomously rather than by a human typing code. That statistic is the real justification for Origin's existence -- Git hosting built decades before agentic coding existed was never architected for that volume or that commit cadence, and Anysphere is betting the gap is wide enough to build a business in.
The product includes automated merge-conflict resolution, stacked pull requests inherited from Cursor's earlier acquisition of Graphite, and metadata tracking that distinguishes agent-written commits from human-written ones -- a feature aimed at engineering managers who want visibility into how much of their codebase is now agent-generated. TheNextWeb reported that Origin is on by default for paid users and that its data-handling terms were not published alongside the launch, a detail that drew immediate scrutiny from developers wary of where their code and commit history now live.
The competitive stakes
GitHub, owned by Microsoft, still hosts the overwhelming majority of the world's private and open-source repositories, and Cursor itself has built its entire product on top of that dominance -- Origin is Anysphere turning its own distribution advantage as the leading AI-native editor into leverage against the company whose infrastructure it has depended on. Pulse previously covered Microsoft's broader AI infrastructure bets, of which GitHub's central role in the developer stack is one of its quieter but most durable assets. GitLab, Bitbucket and a handful of smaller Git hosts have tried and failed to meaningfully dent GitHub's network effects for over a decade; Origin's bet is that agentic coding changes the terms of that competition in a way none of those prior challengers had available to them.
What gets lost in the launch framing is that Origin is a beta, not a mature product, and migrating a team's version control history away from GitHub is a decision most engineering organizations make once a decade, not once a quarter -- inertia, not feature parity, has been the real force protecting GitHub's position for years. The unpublished data terms are the more immediate risk: enterprise engineering teams evaluating Origin will want contractual clarity on code retention and training-data use before they move production repositories, and Anysphere has not yet supplied it.
Cursor's own valuation talks have circled $30 billion-plus this year on the strength of its editor business alone; a credible hosting layer, if it holds up under real production load rather than launch-week benchmarks, gives Anysphere a second product with genuine platform lock-in rather than a feature that can be copied by the next well-funded AI coding startup.
