Back to posts
AINews

Cursor Origin is on for paid plans, and the docs do not say how long your repo is stored

Cursor shipped Origin on August 17, a code host built into the editor, rolling out in early beta to every paid plan unless an enterprise admin opts out. The Origin documentation has a Privacy section, and it says one thing: Origin inherits the namespace owner's Privacy Mode. The words retention, residency, subprocessor and export do not appear on it, and the three Cursor pages that would normally answer those questions were written before Origin existed. Here is what each page actually says today, and how to tell trying it apart from moving.

Cursor shipped Origin on August 17. It is a code host built into the editor: repositories, pull requests, reviews, merges and CI connections in a new Codebase tab, plus a CLI for cloning and pushing. The changelog entry is direct about the rollout. "Cursor can now host your code. Origin begins rolling out today in early beta on all paid plans" — every paid plan user, starting Monday, except enterprise organizations whose admins opt out. Free plans are excluded. Your namespace becomes part of every repository URL, in the shape cursor.com/codebase/acme-corp.

For repositories synced in from GitHub, Cursor is explicit about where authority sits: "Pushes keep going to GitHub, which stays the source of truth for anything started there." Pull request comments sync both directions.

That is the feature. The part worth ten minutes is what happens when you go looking for the storage terms.

What the Privacy section actually covers

The Origin documentation at cursor.com/docs/origin has seven sections, one of which is called Privacy. It contains two statements. The first: "Origin follows the Privacy Mode of the namespace owner: the team or the individual who owns the repo." The second is a prerequisite, that teams on legacy privacy mode have to switch before they can enable Origin.

Search that page for retention, residency, subprocessor, training or export and you get nothing. None of those words are on it.

This matters because Privacy Mode is a training commitment, not a storage commitment. Cursor's own enterprise documentation states the guarantee plainly: "With Privacy Mode enabled your code is never used for training by Cursor or other AI model providers." That is a useful promise and it answers one question. It does not answer how long a hosted repository sits on Cursor's infrastructure, in which region, which subprocessors touch it, or what deleting a repo actually removes. Those are different questions, and a code host is the first Cursor product where they are the load-bearing ones.

The three pages that would normally answer this

Each of these was current before August 17, and each was accurate when it was written.

cursor.com/data-use, last updated July 15, 2026. This is the page Cursor points people to for data handling, and it is thorough about the things it covers: with codebase indexing on, code is uploaded in chunks to compute embeddings, plaintext is deleted after the request, and embeddings plus metadata such as hashes and file names persist in Cursor's database. File contents are cached temporarily under encryption. It names three model providers: SpaceXAI, OpenAI and Anthropic. It does not mention Origin, code hosting or repository storage, because none of that existed when the page was written a month ago.

cursor.com/docs/enterprise/privacy-and-data-governance. This page carries a sentence that was true until last Monday: "Cloud Agents are the only feature that requires Cursor to store code." The surrounding paragraph describes agents running in isolated VMs with encrypted repository copies deleted after the agent completes. The word Origin does not appear anywhere on the page. The same page is precise on residency, so this is not a document that avoids specifics: US-only residency covering inference, processing and storage is available to Enterprise today, EU and Iceland is inference-only on request, and broader EU support is described as in active development. Whether any of that extends to hosted repositories is exactly the sentence that has not been written yet.

cursor.com/privacy, last updated October 6, 2025. The privacy policy predates the product by ten months. On retention it says Anysphere "retains your personal data only for as long as necessary to operate the Service effectively and to support legitimate business needs such as legal compliance, safety, dispute resolution, and enforcement of our agreements." That is standard policy language and it does not convert into a number of days for a git repository. The subprocessor list lives separately at trust.cursor.com/subprocessors.

To be clear about what this is and is not: nothing here shows any code being mishandled. The claim is narrow and checkable. A storage product went live ahead of the pages that describe how the storage works, and it went live switched on rather than switched off.

Trying it and moving to it are different decisions

The distinction Cursor built into the product is the one to use.

A repository synced from GitHub leaves GitHub as the source of truth. Pushes still go there. If you decide against Origin in three weeks, you stop syncing and your history is where it always was. That is a reversible experiment, and it is a reasonable one to run this week if you want to see whether agent-native review is better than what you have.

A repository created in Origin is a different thing. It is the only copy until you make another one, and there is no documented export or migration tooling to lean on. git clone obviously works, so you are not trapped, but "I can clone it" is not the same as a documented answer to what stays behind after you delete the repo.

The timing is the reason to say this today rather than next week. GitHub had a critical incident on the same afternoon Origin launched: opened 13:40 UTC on August 17, resolved 21:15 UTC, seven hours and thirty-five minutes across API requests, Actions, Git operations, issues, pages, pull requests, webhooks and Copilot. Seven of the eight services were declared mitigated at 16:59 UTC and Copilot was not among them. As of this morning there is another open Actions incident at major impact, opened 09:36 UTC. An outage day is exactly when a repository gets moved in frustration, and moving where your source code lives is cheap to postpone by a week and expensive to reverse afterwards.

If you want the resilience conversation rather than the hosting one, the two-tier fallback pattern is the more useful thing to spend the afternoon on. Worth noting from yesterday's numbers: web and API error rates ran around 20 percent, while archive downloads and raw repository content failed at roughly 50 percent. That second layer is the one your CI steps and install scripts quietly depend on, and it does not show up in any dependency file.

Who this lands on differently

Solo developers on Pro. Origin is arriving for you and there is no admin to opt out on your behalf. Nothing moves without you pushing it, so the practical exposure is only whatever you choose to create there.

Team admins. You have an actual decision in admin settings, and it is worth making deliberately rather than discovering later that four repositories were created during a GitHub outage.

Enterprise on legacy privacy mode. You are excluded until you switch Privacy Mode, which means the decision is already staged for you. Use the time.

What to re-check

Cursor will almost certainly publish this. When it does, the four lines to look for are the same four that are missing now: how long a hosted repository is retained, which region it is stored in, which subprocessors touch it, and what a documented export path looks like. The one on the enterprise page to watch specifically is whether "Cloud Agents are the only feature that requires Cursor to store code" gets corrected, since that sentence is now describing a product state that ended on August 17.

The habit underneath all of this is smaller than the news. When a tool you already trust adds storage, the promise you have from it is usually a training promise, and the promise you need is a retention promise. They are written on different pages, and the second one is often the one that has not been updated. Check the date at the top before you rely on what is underneath it.

Ownership context, since it is adjacent and people will ask: SpaceX closed its acquisition of Anysphere on August 14, three days before Origin shipped, and cursor.com/data-use has listed SpaceXAI among its model providers since July 15. We covered the acquisition when it was announced. It is not the reason to wait on Origin. The unpublished retention terms are.


Verified August 18, 2026 against cursor.com/changelog, cursor.com/docs/origin, cursor.com/data-use, cursor.com/docs/enterprise/privacy-and-data-governance and cursor.com/privacy, with incident timings from githubstatus.com. These pages change; the dates above are what they carried at publication.

Get the next post when it ships

One email on Sunday with the new post and a short list of what shipped that week — new guides, tool updates, and a couple of links worth reading.