Cursor Origin: A Git Forge Built for AI Agents, Not Humans

Official Sources

On June 16, 2026, at its Compile conference, Cursor (Anysphere) announced Origin: a Git-compatible code hosting platform whose pitch is unusually blunt. The premise, as the company framed it: what if the primary users of your version control system are no longer human?

That is the whole bet. GitHub, GitLab, and every forge before them were designed around a human cadence: one engineer opens a branch, pushes a few commits an hour, opens a pull request, waits for a review. Cursor's argument is that agentic coding has broken that cadence. When dozens of background agents are committing in parallel, the bottlenecks move, and a forge tuned for humans starts to creak.

This post sticks to what was actually announced, separates the shipped facts from the demo-stage claims, and flags what is still just a waitlist.

Last verified: June 17, 2026.

What Origin Actually Is

Origin is a Git forge: repositories, branches, pull requests, and review, the same primitives you already know. It is Git-compatible, so existing tooling and git push workflows are meant to keep working, and Cursor has talked up GitHub migration tooling to ease the move.

The differentiator is the design center. Where a conventional forge optimizes for human review throughput, Origin is built around agents operating at machine speed: structured, agent-authored diffs, automated review routing for AI-generated PRs, and native integration with Cursor's own background agents.

Crucially, this is not a from-scratch effort. Origin is built on Graphite, the stacked-pull-request platform Cursor acquired, with Graphite co-founder Tomas Reimers leading the demo. That lineage matters: stacked PRs and the review tooling are proven technology, not vaporware. It is the agent-first layer on top that is new.

The Features Worth Tracking

Across the announcement and early coverage, the features that consistently show up:

• Agent-driven merge conflict resolution. A built-in engine aimed at resolving conflicts automatically, the kind that pile up fast when many agents touch the same files concurrently.
• Agent resolution of CI and build failures. Pushing remediation of broken builds toward automation rather than a human triage queue.
• Stacked pull requests. Inherited directly from Graphite, letting a chain of dependent PRs move together, a workflow agents generate naturally.
• MCP- and API-extensibility. Programmatic control surfaces so agents (and your own tooling) can drive the forge directly, not just click through a UI.
• A storage architecture for scale. Coverage describes a hybrid NVMe-plus-S3 design intended to support large numbers of replicas for high-frequency clone and fetch traffic.

Two of these, merge conflict resolution and CI-failure resolution, are the genuinely agent-native ideas. They target failure modes that barely register for a single human developer but become constant friction once a fleet of agents is committing in parallel.

Separate the Numbers From the Hype

Cursor's demo leaned on throughput figures to make the "machine speed" case. The most-quoted is 22.6 commits per second in a single repository, alongside claims of very high clone volumes per hour.

Treat those as demo-stage, vendor-supplied numbers, not independently benchmarked results. They illustrate the design goal, sustaining write and read rates that human-era forges were never built for, but they are staged figures, and some clone-count numbers circulating online have unclear timeframes or units. They are a useful signal of intent. They are not a verified production SLA.

Likewise, treat any third-party rumors of corporate acquisitions tied to the announcement as unverified; they did not come from Cursor's product announcement and are not part of what Origin is.

What Is Real vs. What Is a Promise

This is the most important section for anyone deciding whether to care today.

Real now: the announcement, the product vision, the Graphite foundation, and an open waitlist at cursor.com/origin.

Not yet: general availability. Origin is slated for fall 2026. Pricing has not been disclosed. Enterprise features, security and compliance posture, and self-host options are unannounced. The throughput claims are demo figures awaiting independent verification.

So this is real in the sense that a credible, well-funded company with proven forge technology (Graphite) has committed publicly to shipping it, with a date. It is a promise in the sense that you cannot run your team on it this week.

Why It Matters for Developers

Even as a waitlist, Origin is a useful signal about where agentic development infrastructure is heading.

The honest tension is governance. A forge that auto-resolves merge conflicts and auto-fixes CI failures is removing exactly the friction points where humans currently catch bad changes. That is the value proposition and the risk in the same sentence. The teams that benefit most from agent-speed infrastructure are also the ones with the most to lose if the safety rails are tuned for throughput over correctness. Expect the interesting questions to be about audit trails, approval gates, and who is accountable for an agent-merged change, not about commits per second.

For now, the practical move is small: if you are already deep in the Cursor ecosystem and running background agents at volume, the waitlist costs nothing. Everyone else can watch for the fall release, independent throughput benchmarks, and a pricing page before forming an opinion. The idea is genuinely interesting. The product still has to ship.