OpenAI to acquire Pierre and bring Code Storage Git hosting into Codex
OpenAI has agreed to acquire Pierre Computer Company, Inc., a small developer-tools startup whose Git hosting service and open-source interface libraries already sit underneath parts of Codex. The company announced the deal on November 4, 2026, in a post on its website and on its official newsroom account, saying the acquisition is meant to make Codex "a place where teams can build software together with AI, from first idea to the version they ship." Pierre confirmed the agreement on its own site under the headline "Pierre is being acquired by OpenAI."
According to OpenAI's statement, the plan has two concrete parts. Pierre's paid product, Code Storage, a Git infrastructure service built for AI applications and coding agents, will be integrated into Codex, while Pierre's open-source projects, including the Diffs and Trees libraries that render code changes and file navigation in Codex today, will remain open source after the deal closes. OpenAI said the Pierre team will help it ship new products and platforms.
Financial terms were not disclosed. OpenAI said closing depends on customary regulatory approvals and completion conditions, and that the two companies will continue to operate independently until then. No closing date, purchase price or headcount figure has been published by either company.
The announcement arrives two days after OpenAI described Pando, the Git-compatible backend it built to run its own main code repository, which the company said cut median push latency by nearly 80% once Codex agents began generating far more branches, pushes and fetches than human engineers ever did. Taken together, the two moves show OpenAI building its own source-control layer for agent-written code, a layer that sits squarely in the territory of Microsoft-owned GitHub.
··········
WHAT OPENAI IS BUYING FROM PIERRE COMPUTER COMPANY.
Pierre Computer Company describes itself as a builder of low-level infrastructure "for machines," meaning storage and tooling designed for software agents that create code at a pace and volume that ordinary developer platforms were not built to absorb. The company raised a $22 million Series A led by CRV and 01A in November 2025, according to its own funding announcement, and its customer list on the Code Storage site includes AI app builders and agent companies such as Lovable, Poke, Anything, Bolt, Amp and Parahelp.
The acquisition covers three products that matter for Codex, each with a different status in OpenAI's announcement.
........
Pierre product | What it does | Status stated by OpenAI |
Code Storage | Hosted Git infrastructure for AI apps and agents, with an API and SDKs | To be integrated into Codex |
Diffs (@pierre/diffs) | Open-source library that renders code diffs and files, built on Shiki | Stays open source; already used in Codex |
Trees (@pierre/trees) | Open-source file-tree rendering library with search and drag-and-drop | Stays open source; already used in Codex |
Pierre team | Engineers behind the three products | Joins OpenAI to ship new products and platforms |
........
Diffs is the most visible of the three to developers. Its documentation describes split and stacked diff layouts, Shiki-based syntax highlighting with custom themes, an edit mode with selections and search, virtualization for very large diffs, server-side rendering and worker pools that move highlighting off the browser's main thread, with React and vanilla JavaScript APIs. Trees handles the folder sidebar, with search, drag-and-drop for moving subfolders and markers for files that were recently added or changed. Both libraries ship from Pierre's public GitHub repository, and their own pages describe them as open source without naming a specific license.
Code Storage is the commercial core of the company and the part OpenAI says it will bring inside Codex. Its value for an agent platform lies in letting every generated project, branch and experiment live in a real Git repository that standard Git clients can clone, fetch and push, while the service itself is driven through an API rather than through a human-oriented web interface.
··········
CODE STORAGE SPECIFICATIONS AND PRICING BEFORE THE ACQUISITION.
Pierre introduced Code Storage in October 2025 as a "white-label, ultra low-latency git infrastructure layer" built on a distributed, quorum-based design. The product pages list the following capabilities and limits, which describe the service as it is sold today, before any integration with Codex.
........
Code Storage feature | Published value |
Storage, repositories, API and Git request rates | Listed as unlimited |
Concurrent writes per repository | 30+ |
Maximum repository size | 32 TB |
Observed Git requests per repository | 500+ per second |
Observed API requests | 15,000+ per second |
Availability commitment | 99.99% SLA |
SDKs | TypeScript, Python, Go, plus a CLI |
Sync | GitHub, GitLab, Bitbucket and other Git hosts |
Compliance | SOC 2 Type II, HIPAA |
........
The developer documentation adds features aimed squarely at agents: JWT-based authentication with scoped, expiring tokens and per-client ref policies, Git LFS on the same remote, Git notes, signed commits, ephemeral namespaces for previews and experiments, forks, signed webhooks for push and sync events, and authenticated Git URLs that external sandboxes such as Modal, E2B and Daytona can use directly. The documented workflows include saving and restoring agent memory and agent state in Git, live diffs, and running several parallel attempts at the same task in separate branches.
Pierre's largest public reference is Lovable. In a case study published in August 2026, Pierre said Lovable creates more than 8 million new repositories a week with peaks of 250 repository creations per second, and reported a 230% improvement in p50 fetch latency and a 160% improvement in push latency after the move, with 99.99% uptime for critical repository infrastructure.
Pricing on the Code Storage site, as listed on November 4, 2026, combines a monthly minimum with usage rates.
........
Code Storage plan or rate | Price as listed |
Starter plan | From $20 per month (minimum usage) |
Scale plan | From $200 per month (minimum usage), adds single-tenant option and SSO |
Enterprise plan | Custom, with VPC networking and volume discounts |
Hot storage (touched in last 7 days) | $0.005 per GB per hour |
Cold storage (untouched for 7+ days) | $0.0002 per GB per hour |
Inbound bandwidth (push) | $0.06 per GB |
Outbound bandwidth (clone and fetch) | $0.15 per GB |
........
The hourly storage rates are easier to judge as monthly figures. Over a 730-hour month, hot storage works out to $3.65 per GB and cold storage to about $0.15 per GB, so data that agents keep touching costs 25 times more to hold than data left idle for a week. For an illustrative AI app builder keeping 200 GB of active repositories and 2 TB of older ones, receiving 300 GB of pushes and serving 1.5 TB of clones and fetches in a month, the published rates add up to $730 for hot storage, $292 for cold storage, $18 for inbound traffic and $225 for outbound traffic, or $1,265 per month, comfortably above the Scale plan's $200 minimum. These are list prices for the standalone service; OpenAI has said nothing about whether they will change.
··········
PANDO AND THE GIT BOTTLENECK BEHIND OPENAI'S CODEX PLANS.
The reasoning behind the deal is easier to follow through OpenAI's own engineering account of Pando, published on November 2. OpenAI said that as Codex agents took on more of its internal development, the volume of branches, pushes, merges, fetches and clones in its main repository rose sharply, and its earlier Git systems struggled with slow operations. The company built Pando as a backend that keeps compatibility with the Git protocol and with ordinary Git clients, and said it now runs its main code repository.
The figures OpenAI gave for Pando are a reduction of nearly 80% in median push latency in production and a daily volume of almost a petabyte of repository data. Both describe OpenAI's internal environment and its own workload, and the company presented them as results from that setting rather than as a product benchmark. OpenAI said it is studying workloads in which many agents share a single Git repository, which is the same problem Code Storage was designed to solve for outside customers.
Pando and Code Storage therefore address the same pressure from two sides. Pando is OpenAI's in-house answer for its own monorepo, while Code Storage is a commercial service already used by companies that create millions of repositories for agent-generated projects. OpenAI has not said whether the two systems will be merged, whether Pando will be offered externally, or how Code Storage will be packaged inside Codex.
··········
WHAT CHANGES FOR CODEX USERS AND CODE STORAGE CUSTOMERS.
For people who use Codex in ChatGPT, in the Codex app or through the command line, nothing changes on the day of the announcement. OpenAI said the companies remain independent until closing, and it has not announced a new Codex feature, plan, price or usage limit tied to the deal. The Diffs and Trees components that already render diffs and file trees in Codex continue to work as before. Readers comparing what Codex costs against Claude Code, Cursor and GitHub Copilot can check our guide to AI coding agent pricing across Cursor, Claude Code, Codex and GitHub Copilot, and subscribers following the December changes to Codex quotas can read our report on the shared ChatGPT and Codex token usage limit.
........
Who | What OpenAI has stated | What is still unannounced |
Codex users (ChatGPT plans, app, CLI) | Code Storage will be integrated into Codex | Feature names, plans, limits, release date |
Code Storage customers | Companies operate independently until closing | Future of standalone service, pricing, contracts |
Developers using Diffs or Trees | Projects remain open source after closing | License changes, maintenance roadmap |
Teams on GitHub, GitLab or Bitbucket | No change announced | Whether Codex will offer its own hosted repositories |
........
The most practical consequence of an integrated Code Storage would be repository hosting that lives inside Codex itself, so that an agent's work could be stored, branched, compared and reviewed without first pushing to an outside Git host. OpenAI's wording, which describes Codex as a place to build software "from first idea to the version they ship," points in that direction, though it has not described the product.
Code Storage customers such as AI app builders face the most open questions, since their production repositories run on a service that will soon belong to OpenAI, which also builds competing coding products. Until OpenAI or Pierre publishes terms for existing contracts, these customers have only the statement that the companies operate separately before closing. Teams that depend on the service can review the export options already documented by Pierre, which include exporting repositories and LFS objects to an S3 bucket they own, and its sync features for GitHub, GitLab and Bitbucket.
Developers who embed `@pierre/diffs` or `@pierre/trees` in their own interfaces have OpenAI's commitment that both projects stay open source. That commitment covers availability of the code, and nothing has been said yet about governance, release cadence or how outside contributions will be handled once the maintainers work at OpenAI.
··········
OPENAI, CODEX AND GITHUB AFTER THE PIERRE DEAL.
The acquisition extends Codex from writing and reviewing code into storing it, a layer that has belonged to GitHub, GitLab and Bitbucket for most of the past decade. GitHub is owned by Microsoft, a major OpenAI shareholder and partner, and GitHub's own Copilot coding agent competes directly with Codex for the same developer teams. By adding a Git hosting service designed for agents, OpenAI gains infrastructure that would let Codex keep more of a project's life cycle on its own platform.
The choice of target also reflects how close Pierre already was to Codex. Its Diffs and Trees libraries render the code changes and file sidebars that Codex users review every day, so OpenAI is buying a team whose work is already part of its product, together with a storage service that has been tested at the scale of AI app builders creating millions of repositories a week.
For GitHub, the immediate effect is limited, because Code Storage is an infrastructure product sold to platforms rather than a social coding site with pull requests, issues and a public project community. The longer-term question is whether OpenAI will expose hosted repositories directly to Codex users, which would give developers a first alternative to keeping agent-written code on a third-party host.
··········
OPENAI'S PIERRE ACQUISITION: CONFIRMED FACTS AND OPEN QUESTIONS.
OpenAI has confirmed that it will acquire Pierre Computer Company, that Code Storage will be integrated into Codex, that the Diffs and Trees projects will remain open source, that the Pierre team will help it ship new products and platforms, and that the deal still needs customary regulatory approvals, with both companies operating independently until it closes. Pierre has confirmed the agreement on its own site. The product facts in this report, from Code Storage's 32 TB repository limit to its published storage and bandwidth rates, come from Pierre's own pages as they stood on November 4, 2026, before any integration.
Several points remain undisclosed: the purchase price, the expected closing date, the fate of Code Storage as a standalone service and of its current customer contracts and prices, the license and governance of the open-source libraries after closing, and the form Code Storage will take inside Codex, including which ChatGPT plans would get it and with what limits. OpenAI has also not said whether its internal Pando backend will become part of the same offering.
What is already clear is the direction of OpenAI's coding business. With Pando inside the company and Code Storage on the way, OpenAI now controls both an internal Git backend tuned for agent traffic and a commercial one already proven at millions of repositories a week, and Codex is positioned to manage where agent-written code lives as well as how it is written.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org
