# Cursor Origin Puts Git Hosting Inside the Agent Workflow. Migration Can Wait.

**Cursor Origin entered early beta for all paid plans on August 17, 2026, but teams should test it with a mirrored repository before moving their source of truth.** Cursor has confirmed GitHub repository sync and positioned Origin as a forge built for agents. It has not published a reproducible benchmark method, a migration guide, or Origin-specific governance documentation. The decision should turn on the full change cycle from task assignment to reviewed merge, not commit throughput alone.

Origin extends Cursor beyond the editor. Once coding agents create changes continuously, the forge becomes part of the execution path. It handles identity, repository state, review gates, conflict resolution, and the evidence attached to a merge. Cursor now has an opportunity to connect its editor, Cloud Agents, Bugbot, and hosting under one operating model.

That opportunity deserves a pilot. It does not yet justify a migration.

![Cursor Origin official landing page showing early beta access for paid plans.](https://s4.tenten.co/learning/content/images/2026/08/landing-page-1-9.png)

#### What Cursor Has Confirmed

The official Origin page calls the product “A git forge for the agentic era.” It also says early beta is available on every paid plan. Cursor's launch announcement adds one concrete onboarding path: users can sync repositories from GitHub. The sync approach matters because it lets a team test Origin without making an immediate cutover.

The launch demo emphasized high-volume commits, clones, and global synchronization. Cursor's product page does not disclose repository size, object count, region layout, concurrency, test duration, or a reproduction method. A team should therefore build its own baseline instead of using the demo to forecast production results.

| Decision point | Public evidence | What a team still needs to test |
|---|---|---|
| Availability | Early beta launched August 17, 2026 for all paid plans | Organization and region eligibility |
| GitHub connection | Repository sync is supported | Bidirectional sync, force pushes, LFS, and submodules |
| Commit throughput | No reproducible public benchmark | A 30-minute run on the team's own monorepo |
| Clone throughput | No reproducible public benchmark | Cold-cache and peak-concurrency behavior |
| Global sync | No reproducible public benchmark | p50, p95, and p99 across operating regions |
| Governance | No Origin-specific public guide | Audit logs, retention, restore, and exit procedures |

An early beta can reasonably arrive before its full documentation. Source control still has a higher proof threshold than an editor extension. It is often the recovery layer when the rest of the delivery system breaks.

#### The Graphite Acquisition Explains the Direction

Cursor announced a definitive agreement to acquire Graphite on December 19, 2025. At the time, Cursor said Graphite's code review platform served hundreds of thousands of engineers. The announcement identified review, safe merges, and collaboration as the bottlenecks that appear when code generation gets faster.

Origin follows a visible product sequence. Cursor first owned the coding surface. It added Cloud Agents and Bugbot, then acquired a team with deep review experience. Git hosting now brings repository state into the same product boundary.

The appealing end state is a shared context for writing code, opening a review, responding to feedback, and completing a merge. The harder problem moves downstream. Human reviewers need to understand the added change volume. CI must absorb more runs. Access mistakes need fast containment. Throughput at the Git protocol layer cannot answer those questions.

#### GitHub Is Already Adapting Its Workflow

The competitive frame is more interesting than a new forge attacking a static incumbent. On March 24, 2026, GitHub added the ability to mention Copilot inside a pull request and ask it to change code. The agent runs in a cloud development environment, executes tests and linters, and pushes the result.

On April 1, GitHub let Copilot cloud agent work on a branch before opening a pull request. Users can review a plan and diff before requesting review. On June 18, Copilot code review gained support for a repository-level `AGENTS.md` file.

GitHub is adding agent behavior inside an established collaboration system. Cursor is expanding outward from the agent's workspace. Both are competing for the same control point: the system that decides when an agent may change code, when a human must intervene, and what evidence is required before a merge.

#### Bundling Lowers the Cost of a Pilot

Cursor has not listed a separate Origin price. Its current Individual plan starts at $20 per month. Teams Standard costs $40 per user per month, while Enterprise uses custom pricing. Including the beta with every paid plan gives Cursor a large test population without requiring a second purchase decision.

The bundle leaves two commercial questions open. Cursor has not said whether the production service will charge for storage, transfer, agent activity, or active users. It has also not specified how existing Enterprise controls such as audit logs, repository access controls, and service accounts apply to Origin.

Cursor's security page was updated on April 24, 2026. It lists SOC 2 Type II, annual penetration testing, Privacy Mode, and Enterprise administration controls. Those facts support a vendor review. They do not replace product-level answers about data residency, backup targets, incident recovery, Git hook compatibility, and repository export.

#### A Pilot That Protects the Main Repository

I would choose one noncritical repository and keep GitHub as the source of truth. Origin can receive a mirror while the team runs normal daytime work and scheduled agent jobs for at least two weeks. Clone latency belongs in the record, but it should not dominate it.

| Metric | Measurement | Evidence required before migration |
|---|---|---|
| Agent change cycle | Median and p95 from assignment to merge | Lower cycle time without a higher failure rate |
| Review load | Human minutes per pull request | Stable readability as change volume rises |
| CI pressure | Daily runs, queue time, and cost | Agent activity does not convert saved time into CI delay |
| Sync correctness | Daily commit graph and tag comparison | Zero loss and zero unexplained divergence |
| Recovery | Restore drill after an intentional branch deletion | RPO and RTO meet the team's policy |

If the pilot proves only that commits arrive faster, keep the existing forge. Origin needs to reduce review delay and CI wait while preserving recovery confidence.

#### FAQ

##### What is Cursor Origin?

Cursor Origin is a Git hosting product designed around agent workflows. Its early beta opened to all paid Cursor plans on August 17, 2026, and it can sync repositories from GitHub.

##### Can Cursor Origin replace GitHub today?

The public evidence does not support a full migration yet. Teams can use a noncritical mirrored repository to test sync correctness, review load, CI cost, permissions, and recovery before changing their source of truth.

##### What public performance evidence is available?

Cursor has shown launch-demo performance, but it has not published the dataset, infrastructure, concurrency model, or reproduction method. Teams need to measure their own repositories and operating regions before forecasting production results.

##### Which Cursor plans include Origin?

Cursor says the early beta is available on all paid plans. Individual pricing starts at $20 per month, Teams Standard costs $40 per user per month, and Enterprise pricing is custom. Production pricing for Origin has not been published.

#### Sources

- [Cursor Origin product page](https://cursor.com/origin)
- [Cursor launch announcement](https://x.com/cursor_ai/status/2089399057659596847)
- [Cursor announcement that Graphite is joining the company](https://cursor.com/blog/graphite)
- [Cursor pricing](https://cursor.com/pricing)
- [Cursor security](https://cursor.com/security)
- [GitHub Copilot cloud agent update](https://github.blog/changelog/2026-04-01-research-plan-and-code-with-copilot-cloud-agent/)
- [GitHub pull request agent update](https://github.blog/changelog/2026-03-24-ask-copilot-to-make-changes-to-any-pull-request/)
- [GitHub Copilot code review support for `AGENTS.md`](https://github.blog/changelog/2026-06-18-copilot-code-review-agents-md-support-and-ui-improvements/)

#### Author Insight

I see Origin as part of the agent runtime, not a new skin for GitHub. That makes the product ambitious and raises its reliability bar. When agents keep sending changes after the team goes home, a reproducible audit trail and a proven exit path matter more than a polished throughput chart.