Cursor unveils Origin, an agent-first Git host; GitLab and Zed take other paths

For the first time since GitHub set the default shape of CI/CD ingress, three vendors have publicly said the default needs replacing, and at least one of them has put a name on the candidate. Graphite co-founder Tomas Reimers unveiled Origin, described as a Git-compatible code-hosting platform built from the ground up for a world where AI agents do the bulk of the work, at an invite-only Cursor developer conference in San Francisco, The New Stack reported on 18 June 2026. Cursor acquired Graphite in a deal announced in December and closed in January. GitLab and Zed, per the same report, are pursuing different rebuilds of the GitHub model for the same agent-native world.

What landed on the slide

Origin sits inside Cursor; Graphite is now part of Cursor; SpaceX has separately agreed to acquire Cursor in an all-stock deal valued at $60 billion, per The New Stack. GitLab and Zed were named alongside Origin as taking their own paths to the same problem, though the reporting stops short of saying they ship the same shape of system.

What none of the three vendors has yet said in this reporting is what any of these hosts actually changes at the wire level. There is no new webhook payload published. There is no new review-bot identity model on the table. There is no public branch-protection schema. There is a thesis, that the GitHub mould is wrong for agentic work, and a product name.

Why the host choice is a CI/CD concern

Every CI pipeline on the planet starts with a contract its developers rarely read. The Git host decides which event arrives at the runner, what user object signed the push, which check runs can post statuses back, and which secrets a workflow is allowed to see. That contract was written for humans pushing branches and reviewing pull requests on roughly human timescales.

Agentic work does not honour those timescales. Many small pull requests, opened in a batch, asking for reviews from each other, producing diffs that pass linters and may or may not match intent. The operational consequence is that webhook semantics, PR reviewer identity, branch protection rules and check-run permissions stop being implementation details and start being the API surface the pipeline depends on. If a new host decides a review signed by an agent identity does not satisfy the same branch-protection rule as a review by a human, every required-reviewer policy ever written against GitHub has to be reread. If it gives bot pushes their own ref namespace, every on: push workflow you ship has to be reread too.

Three rebuilds, three opinions

The reporting leaves the three approaches roughly sketched. Origin is the named product, launched from inside Cursor. GitLab and Zed are pursuing rebuilds of the same model, the report does not detail what either of them ships, only that they disagree with each other and with the Cursor camp about the right shape.

For practitioners that is the headline: even the people convinced GitHub needs replacing cannot yet agree on what to replace it with. A pipeline that has to be portable across three competing agent-host opinions ends up written against the lowest common denominator, which today is still the GitHub API.

What to watch from the pager

Two things. Whether any of these hosts publishes the actual events, identities and protection primitives that distinguish them from GitHub's, the names of webhook payloads, the shape of the bot-identity object, what a branch-protection rule looks like when one of its required reviewers is an agent. And whether any major CI provider commits to a first-class integration on a host other than GitHub before customers commit to one. Until then the rebuild story is a slide. The pipeline still runs on the old host.