Cursor Unveils Origin Code Hosting Platform: Rebuilding Code Infrastructure for the AI Agent Era

In the fiercely competitive landscape of AI programming tools, Cursor has dropped a bombshell. On June 16, at its inaugural Compile conference in San Francisco, Cursor announced Origin, a code storage and Git hosting platform. This is not an ordinary product expansion—it is a definitive answer to the fundamental question: what should code infrastructure look like in the AI era?

Origin's core design philosophy is embodied in one critical decision: it treats AI agents as first-class citizens, not as subordinate roles to human developers. Traditional Git platforms—whether GitHub, GitLab, or Bitbucket—were all designed, without exception, for human developers. The typical behavioral pattern of a human developer involves a handful to a few dozen commits per day, occasional code reviews, and periodic cloning and pushing. But when AI agents enter the development workflow, workload patterns change fundamentally.

Cursor CEO Michael Truell pointed out at the conference that agents clone, read, and push at frequencies orders of magnitude higher than human developers. More critically, multiple AI agents may work simultaneously on different branches of the same repository, making merge conflicts not an occasional event but a systematic norm. Origin was built specifically to address this entirely new challenge.

Origin's underlying technology stack comes from Graphite, which Cursor acquired in December 2025. Graphite was originally a developer tool focused on high-performance code review and stacked pull requests, with its core technical expertise centered on handling large volumes of concurrent branches and merge conflicts. Origin re-architected Graphite's technical foundation, with targeted optimizations for scenarios where AI agents work in parallel.

On the main screen at Compile, Cursor displayed a striking set of performance metrics: approximately 296,000 clones per hour, roughly 81,000 pushes, and 22.6 commits per second on a single repository. Global synchronization latency is under 400 milliseconds, with automatic failover completed within 10 milliseconds. The storage layer is built on S3, supporting unlimited replica scaling. The significance of these numbers lies not in their sheer magnitude but in the design constraints they represent: Origin was built on the assumption that AI agents are the primary traffic source, rather than patching agent support onto a human-centric workflow.

Automated merge conflict resolution is Origin's most radical differentiating feature. In traditional Git workflows, when two developers edit adjacent sections of the same file and create a conflict, human intervention is required to decide whose changes to keep. But in scenarios where agents work in parallel, conflicts can appear across dozens of files simultaneously, making manual resolution entirely impractical. Origin includes an AI-powered conflict resolution engine that can automatically analyze conflict context, assess each agent's modification intent, and complete merges without human intervention in the vast majority of cases.

The timing of Origin's launch carries clear competitive strategic intent. GitHub has experienced a series of reliability and security incidents in recent months, with many enterprise customers already publicly or privately seeking alternatives. While GitHub Copilot is a pioneer among AI coding assistants, its underlying code hosting infrastructure remains a design from a decade ago—its repository model, permission systems, and CI/CD integration are fundamentally optimized for human team workflows and lack native support for the "first-class citizen" needs of AI agents. Cursor is exploiting this window: at a time when GitHub remains the dominant platform, it is launching a hosting layer designed from the ground up for AI agents.

Origin was not the only product Cursor announced at Compile, but it is the most strategically significant one. The conference also featured two other major announcements: SpaceX's acquisition of Cursor in a $60 billion all-stock transaction, and a new 1.5-trillion-parameter model being trained from scratch on over 100,000 GPUs on xAI's Colossus supercomputer. These two announcements give Origin broader strategic depth.

The tie-up with SpaceX means Cursor gains access to far-earth orbit data centers and global low-earth-orbit satellite internet infrastructure, which, combined with Origin's "sub-400-millisecond global synchronization," can form a complete technology stack spanning from the physical layer to the application layer. Meanwhile, the ultra-large model collaboration with xAI suggests that Origin's built-in AI conflict resolution and code review capabilities will be powered by an inference engine far stronger than any current coding model.

On the model front, Truell disclosed the new model plan in relatively complete detail for the first time. This model, with over 1.5 trillion parameters, is being pre-trained on more than 100,000 GPUs, at a scale comparable to Opus and GPT. Truell even revealed that Opus 4.5 through 4.8 and GPT-5 through GPT-5.5 are actually all under 2 trillion parameters. He believes the only moat at this stage is scaling.

Compared to models Cursor has used in the past, the new model has three notable changes: it is significantly larger, approaching the scale of frontier AI labs; it is trained from scratch rather than fine-tuned from open-source models, giving the team more complete control over model behavior; and the compute used for training is far beyond previous efforts, roughly a 10- to 20-fold increase. Truell emphasized that this new model will not be limited to coding capabilities—the team wants it to possess more general intelligence, because Cursor's goal is not just to have AI write code, but to enable anyone to build what they want.

Cursor Mobile also made its debut at the conference, allowing users to launch and control AI agents directly from their phones, including sending tasks and viewing editing results. In an agent workflow—where one or more AI agents are continuously creating branches, committing code, and running tests in the background—the ability to monitor and intervene in real time from a mobile device becomes a necessity. The implicit premise of this workflow design is that most developers in the future will not write code directly, but will manage and supervise the coding output of AI agents.

From an industry landscape perspective, Origin's launch marks a systematic shift of AI coding tools from the editor layer down to the infrastructure layer. Cursor began as an editor, later evolved into a standalone AI IDE, and is now building its own code hosting and collaboration layer. This vertically integrated path stands in sharp contrast to GitHub Copilot's horizontal integration approach. The advantage of the vertical integration route is control and optimization depth—Origin can assume that its users are also users of the Cursor editor and Cursor Agent, thereby establishing a closed but highly efficient feedback loop between editor, agent, and repository. The disadvantage is network effects—GitHub has over 100 million developer users, and the vast majority of open-source project communities are deeply embedded in its ecosystem.

In his speech, Truell also shared his views on other leading AI labs. In his assessment, Google has not pushed model scale to comparable levels, while also pursuing sparsification too aggressively, and its post-training and reinforcement learning capabilities have not yet been fully translated into a stable, user-friendly product experience. OpenAI is still digesting the pressure of route adjustments following GPT-4.5. He views Anthropic as the only AI lab that has successfully made the leap to roughly 10-trillion-parameter scale, and believes it will not be easy for OpenAI to catch back up before year-end.

Origin is expected to officially launch in fall 2026, with a waitlist now open for registration at cursor.com. Pricing and detailed enterprise features have not yet been disclosed. Cursor has committed to providing API and MCP protocol support, making it easy for users to integrate Origin into existing CI/CD pipelines.

For the developer community, Origin's emergence raises a fundamental question: when AI agents become the primary source of code contributions, does our collaboration infrastructure need a complete overhaul? Cursor's answer is an unequivocal yes, and Origin is the concrete realization of that answer.