If your product keeps its code on Pierre's Code Storage, map your exit this month: OpenAI has said you can stay after closing, and it has not said for how long. OpenAI announced on November 4 that it will acquire Pierre Computer Company and move Code Storage's features into Codex, with no shutdown date given. The price was not disclosed.
If you buy Codex or GitHub and do not use Code Storage, leave your plans alone this quarter. The deal has not closed, and the announcement does not say whether Codex will offer a public Git hosting product.
What OpenAI Is Buying From Pierre
OpenAI is buying a Git storage service built for machine traffic, plus open-source developer libraries that Codex already runs. Code Storage exposes native Git clone, push and fetch endpoints, SDKs in TypeScript, Python and Go, and webhooks. Pierre's own page lists a 99.99% SLA and a 32TB repository limit. Those are vendor figures.
The open-source side is already inside the product. OpenAI's post, as quoted by Cellcog, says anyone who has browsed files or reviewed a change in the Codex app has used Pierre's Trees and Diffs libraries. OpenAI also says Pierre's open-source projects will stay open source and that it will keep contributing to them.
Closing depends on customary conditions including regulatory approvals, and the two companies stay separate until then. No closing timetable was given.
One sourcing caveat changes what you should do. OpenAI's post and the Axios exclusive both refused our automated fetch on November 4, 2026, so the OpenAI wording here comes from outlets quoting it. Have your counsel read the original before relying on any sentence of it.
Why Code Storage Customers Face a Competitor Risk
Some Code Storage customers would be storing their product's core asset with a company that sells against them. Pierre names Lovable, Poke, Anything, Bolt, Amp and Parahelp as customers, and Amp describes itself as an AI coding tool with remote agents, the same category as Codex.
The dependency runs deep for at least one of them. Pierre's August case study on Lovable reports more than 8 million new repositories a week and a peak of 250 repository creations a second, and it quotes Lovable's Will Rudenmalm: "Pierre has become foundational infrastructure for how we operate AI-native software generation at scale." Those are Pierre's published numbers. Lovable's head of growth, Elena Verna, told the 20VC podcast earlier this year that she worries more about "OpenAIs, Anthropics, Googles, Apples" than about smaller rivals. That comment predates this deal, and we found no statement on the deal from Lovable, Bolt or Amp as of publication.
The strongest case for staying put is a good one. OpenAI's stated plan is to let existing customers continue using the service, nothing published sets an end date, and a migration now spends engineering time on a risk that may never arrive. So hold off on moving. The exit plan is cheap compared with a migration, and you can write it this month without changing production.
What the Code Storage Terms Allow
Pierre's public terms give you little protection if the plan changes after closing. The standard agreement, last updated October 27, 2025 and checked November 4, 2026, says "either Party may terminate this Agreement, at any time and for any or no reason, upon written notice," effective on the last day of the month the notice is received. A notice received on November 28 would take effect on November 30. Your order form may say otherwise, which is why it is the first document to pull.
Three other gaps are worth reading for yourself:
- Either party can assign the agreement without consent to a successor that acquires substantially all the relevant assets, and there is no separate change-of-control clause.
- The agreement does not spell out return or export of Customer Data after termination, though it says you keep your rights in that data.
- The data processing addendum promises to "permit Customer to delete or obtain copies of Personal Data" after termination, with no deadline or format, and that covers personal data only.
OpenAI has shut down an acquired customer service before. After it bought Rockset in 2024, Rockset's FAQ gave customers paying month-to-month or in arrears until September 30, 2024 to off-board. We could not find the terms for customers on contracts. Hydrolix's Christina Cewe, whose company was chasing those customers, criticized the "only 90 days of notice", so discount the tone and keep the date. Rockset is one precedent, and this time OpenAI has said customers can continue. Our coverage of Apple's Ascend.io deal shows the same migration problem at another buyer.
How Far Two-Way Sync Gets You Out
Sync gives you a cheap copy of branches and tags, and it leaves some of your data behind. The sync modes differ in ways that matter to an exit plan:
| Mode | Direction | Git LFS | What stays behind |
|---|---|---|---|
| GitHub App | Two-way, with GitHub as source of truth | Yes | Ephemeral refs |
| Public GitHub | One-time import, no ongoing sync | No | Everything after the import |
| Generic HTTPS (GitLab, Bitbucket, Gitea, Forgejo and others) | Copies to and from the host | No | Ephemeral refs and LFS objects |
Ephemeral refs are the gap most likely to bite an agent product. Pierre's docs recommend them for agent snapshots, preview builds and scratch commits, and they never reach the upstream host. The docs also say sync mirrors default-namespace branches and tags, so test anything you keep elsewhere, including the custom Git notes refs Pierre added in September.
GitHub App mode has a second cost. Code Storage sends all pushes directly to GitHub, so turning it on for every repository routes your write path through the host you may have left because of load. A platform creating repositories at Lovable's reported rate will want to sync selectively and archive the rest.
The docs describe bulk imports, and we found no bulk export tool. A Git Archive endpoint downloads a tar.gz of one ref or commit, pitched at packaging docs and release artifacts. For a full copy, a normal clone is your export, and the list prices tell you what it costs. Outbound bandwidth is $0.15 per GB, so cloning 10 TB out costs about $1,500 by our arithmetic. Watch the storage line too. Cold storage is $0.0002 per GB per hour and hot storage, for data touched in the last seven days, is $0.005. Pierre's docs say cold repositories thaw automatically on read. If a clone counts as a touch, seven days of 10 TB at the hot rate is $8,400 against $336 at the cold rate. Ask Pierre before you run the job, and stagger it if the answer is yes.
Your rewrite cost sits above Git. Native clone and push will work against any host, while calls to Pierre's SDKs, its webhooks and its ephemeral remote will need replacing.
Should Codex Buyers Change Their GitHub Plans?
No. Treat a Git backend inside Codex as an unshipped roadmap item. The Axios headline, "OpenAI targets Microsoft's GitHub," is Axios's framing, and OpenAI's announcement does not mention GitHub. Microsoft owns GitHub and nearly a quarter of OpenAI, which makes any direct fight awkward to predict.
The evidence that OpenAI can run Git at scale is real and self-reported. Its Pando backend, described on November 2, supports all development in its monorepo, cut median push latency by nearly 80% and delivers nearly a petabyte of Git data a day. OpenAI's earlier systems at times fell below 90% reliability under agent load. Those are company figures for one company's repository. OpenAI is exploring whether to make the technology available to customers, and whether Pando and Code Storage merge or stay separate is an open question.
You do have more options than a year ago, and none is a like-for-like enterprise replacement yet. Cursor launched Origin on August 17, with GitHub sync in both directions, on the day of GitHub's outage. Thomas Dohmke's Entire unveiled its Git forge on October 26, claiming pushes up to 89 times faster than GitHub, with large-enterprise offerings and self-hosting still planned. Meanwhile GitHub has changed your renewal math on its own: GitHub Enterprise Cloud bought through an account team now includes Copilot access at no seat charge, with usage billed from the first request. That is the number to model at renewal.
OpenAI is also assembling the pieces. Ona added a place for Codex agents to keep running, and Pierre adds a place for their output to live. Add source control to your 2027 Codex questions, and ask for a date.
What to Do Before the Deal Closes
This Week:
- Pull your Code Storage order form and find its termination, notice and data-return language. If you are on the standard terms, tell your general counsel about the month-end clause.
- Inventory your dependency. Count repositories, total GB, LFS use, ephemeral refs, notes refs, and every SDK and webhook call site, and name one engineer as owner.
- Email your Pierre contact three questions: how long the service will run after closing, whether Managed Code Storage in your own cloud continues, and whether OpenAI gets any access to customer repositories or usage data.
This Month:
- Clone a sample of cold and hot repositories to a second host, then check branches, tags, LFS objects and notes refs against the source. Record the bill.
- Turn on sync for the repositories you could not rebuild, such as paying customers' projects. Use a host you already run, and confirm LFS coverage first if that host needs Generic HTTPS.
- Put the Pierre SDK behind a thin internal interface in the next code you touch there, so a future swap has one call site.
Before Closing:
- Ask for written commitments in your order form: a minimum notice period before end of service, an export window, a price hold and a no-training clause for your repositories. A closing is your last natural moment to ask while Pierre is still a separate company.
- Write down the trigger that starts a migration. Good ones are a published end date, a price change, or a contract answer you do not accept.
The Bottom Line
OpenAI is folding a neutral infrastructure supplier into a product that competes with some of that supplier's customers, and its promise so far covers continued use with no term attached. Rockset's month-to-month customers got about three months in 2024. Plan for that length of notice unless your contract says longer, and run the clone test this week so you know what leaving costs.
Continue Reading
- OpenAI Buys Ona: Codex Agents Now Run for Days, Not Minutes
- GitHub Enterprise Adds Copilot, Bills From First Use
- Copilot Retried Into GitHub's Outage. Cap Your Agents.
- Why SpaceX's $60B Cursor Deal Ends Enterprise AI Neutrality
- Your AI Vendor Joined OpenAI in March. You Heard in August.
- The Pinned Commit Was a Branch. Check Your Git Host.