Anyone who has browsed files or reviewed a change in the Codex app has already used Pierre's work. OpenAI agreed on November 4 to acquire Pierre Computer Company, the startup behind the Code Storage service and the open-source Trees and Diffs libraries that render file navigation and change review inside Codex. OpenAI did not disclose terms. The deal still needs regulatory approval, the two companies operate independently until it closes, and Code Storage customers can keep using the service while OpenAI brings its features into Codex.

The timing gives the deal its context. Two days earlier, OpenAI published an engineering post on Pando, the in-house source control backend that now runs all development in its monorepo. Quarterly commits to that monorepo will have grown by more than 2,400% from the first quarter of 2024 to the end of this year (OpenAI, 2026), and before Pando the reliability of OpenAI's Git backend had slipped below 90%, with slow fetches and failed merges (OpenAI, 2026). One engineer designed Pando's core, teams of Codex agents implemented it in Rust over 10 weeks, and it went into production in April. Pando is internal. OpenAI says it is looking at ways to share the underlying technology but has set no date or format.

Pierre is the commercial half of that picture. Jacob Thornton, a co-creator of Bootstrap, and Ian Ownbey founded the company in 2023 out of Y Combinator, and Code Storage sells hosted Git with interfaces built for software that writes code, not only for people pushing commits.

OpenAI Kept Git and Rebuilt the Server Underneath It

A branch in Git is a pointer to its latest commit, and moving that pointer is the only part of a push that needs coordination. Conventional Git servers order those updates across the entire repository, so when dozens of agents work in parallel, one agent's push to its own branch can sit waiting behind another's. Pando handles each branch on its own, streams incoming code into blob storage, and serves downloads from regional caches. OpenAI names the tradeoff openly: a client reading several branches at once may see them at slightly different moments, which OpenAI argues matters little because people and agents care about the order of updates on one branch at a time.

Code Storage takes the same posture toward compatibility, which is why the two fit together.

Pierre's service speaks standard Git and adds machine-oriented application programming interfaces on top. Its published specifications cover repositories up to 32 terabytes and at least 30 concurrent writes (Blockchain.News, 2026), along with Git Large File Storage, software development kits for TypeScript, Python and Go, and synchronization with GitHub, GitLab and Bitbucket. That last feature carries the most weight for an enterprise buyer, because it lets a team give agents their own high-throughput storage without abandoning the forge where its branch protections, approvals and audit history already live.

Every Major Coding Vendor Is Racing to Own Version Control

OpenAI is one of several companies rebuilding this layer at once. GitHub devoted much of its recent Universe keynote to making its own infrastructure faster and more reliable for agent-scale development. Former GitHub chief executive Thomas Dohmke started Entire around Git for agents, Cursor launched a project called Origin, and Anthropic went the other direction with Scion, a replacement for Git with its own client and architecture that it has said it will not release. OpenAI chose to stay inside Git and contribute many of its client fixes back to the open-source project.

The Pierre deal also extends a pattern. In March, OpenAI announced plans to buy Astral, maker of the Python tools uv and Ruff, and in June it agreed to acquire Ona, which runs persistent cloud environments where agents keep working after a laptop session ends. Each purchase moves OpenAI closer to the parts of the software toolchain that sit around the model, and SiliconANGLE noted that observers read the Pierre deal as a move toward territory GitHub holds today. GitHub belongs to Microsoft, one of OpenAI's largest backers, so the partner relationship and the competitive one now run through the same repository.

The same day OpenAI announced Pierre, Meta bought Conductor, the app engineers use to run several coding agents side by side. The two announcements set different terms for existing customers. Conductor published a shutdown schedule, with its cloud service going read-only on November 13 and closing on November 20. OpenAI committed that Code Storage customers can keep using the service after the deal closes while its features move into Codex. Neither announcement explains the buyer's motive, and you do not need it to. The written product commitment attached to the deal is the part you can plan around.

When agents write most of the commits, the forge you choose becomes the place you enforce governance.

Open Libraries and a Paid Storage Service Follow Different Paths After Close

OpenAI committed that Pierre's open-source projects will remain open source and that it will keep contributing to them. Trees, Diffs and the newer Highlights syntax highlighter are interface components, and keeping them open costs OpenAI little while encouraging other tool makers to build on the same review experience Codex uses.

Code Storage is a different asset. It is a hosted, paid service, and OpenAI's plan is to move its features into Codex. That plan is reasonable for a company whose agents generate code faster than conventional forges were built to absorb, and it also means the storage your agents write to could become part of the same contract as the agent itself. You gain one throat to choke. You also concentrate the code, the history and the review trail with a single vendor whose priorities include its own internal monorepo.

The review trail is where enterprises hesitate most. When I looked at GitHub's governance features for Copilot agents in April, my argument was that enterprises hesitate less over code quality than over whether they can explain what an agent did, when, and with which model. If Codex becomes a place agent code is stored as well as written, OpenAI will need to answer that same audit question for its own storage layer.

Unknowns and Uncertainties

• OpenAI has not disclosed a purchase price, an expected closing date or an integration timeline for Code Storage features in Codex.
• Pricing and account terms for existing Code Storage customers after close have not been announced.
• OpenAI has not said whether Code Storage's synchronization with GitHub, GitLab and Bitbucket will carry into the Codex version.
• Pando and Code Storage are separate systems with different architectures, and OpenAI has not said how, or whether, the two will combine.