What did OpenAI agree to buy?
OpenAI said on 4 November 2026 that it will acquire Pierre Computer Company, Inc. The startup builds Code Storage, a paid Git-based storage service for machines, and maintains several open-source tools that OpenAI already uses. The announcement appeared on OpenAI's company blog under the line “Craftsmanship meets scale for the agentic era”.
Pierre's libraries are not new to Codex. OpenAI says the Trees library powers file navigation in the Codex app, while Diffs renders code changes for review. Diffs also has an edit mode that lets a reviewer fix a line inside the diff. Pierre recently released Highlights, a syntax highlighter that updates as a developer types or an agent streams code.
Code Storage is the commercial product. Pierre describes it as a distributed storage system built on Git, with “API ergonomics optimized for machines”. Companies use it to store generated code, agent memory files, design artifacts and other documents that change over time.
Financial terms were not disclosed. SiliconANGLE reported that the deal value was unknown at announcement. OpenAI also did not say how many Pierre employees would join, or how the two engineering teams will be combined.
The strategic logic starts with Git. Git records every change to a codebase, and almost every professional team depends on it. When agents write more code in parallel, the repository has to absorb more branches, more pushes and more reads. OpenAI is buying a team that has already built tooling for that load.
The transaction is not complete. OpenAI says closing is subject to customary conditions, including regulatory approvals, and that the companies remain separate until then. Pierre's open-source projects will stay open source, and customers can keep using Code Storage after closing while OpenAI moves its features into Codex.
Why does Git struggle with AI coding agents?
Git was designed for humans exchanging patches at human pace. AI agents change the volume and concurrency of that work. OpenAI explained the problem in detail two days before the acquisition, in an engineering post titled “Scaling Git at OpenAI”.
The post says Codex lets engineers do far more work in parallel. That creates more branches, more pushes, more merges and more fetches from development environments, research jobs and CI systems. The load compounds because one engineer may have several changes ready for CI at once, and each push can trigger several jobs fetching the same commit.
OpenAI's monorepo began to buckle. Engineers saw slow fetches and pushes, failed merges, and reliability that fell below 90% at times. The company had to decide whether it had hit a limit in Git itself or in the way it served its repository.
Its answer was Pando, a new backend for Git repositories that keeps the Git protocol and client compatible. Pando now powers all development in OpenAI's monorepo. OpenAI built it with Codex in 10 weeks, and it began serving production traffic in April 2026.
The results are internal production figures, not third-party benchmarks. OpenAI says Pando cut median push latency by nearly 5x, or about 80%, and now delivers nearly a petabyte of Git data per day. It also says Pando served more than 57 terabytes of Git packs in a single hour, with over 99% of those bytes coming from local disk cache.
The design explains the gains. Pando coordinates updates to each branch independently, so pushes to unrelated branches can proceed in parallel. It also reuses packfiles across fetches instead of building a separate pack for every client. The tradeoff is causal consistency: a reader looking at several branches can see them at different points in time.
OpenAI considered two common ways to scale Git and rejected both for its workload. One coordinates every reference update across replicas with three-phase commit, which puts disk and network latency on the push path. The other orders updates through a shared write-ahead log in blob storage, which makes unrelated pushes compete for the same manifest. Pando avoids both bottlenecks by giving each branch its own update path.
That choice has consequences for tooling. A build system that reads several branches may not see a single snapshot of the repository. OpenAI accepts that tradeoff to keep pushes fast. Code Storage makes the opposite bet, promising consistent reads and writes across replicas.
What is Code Storage, and who already uses it?
Code Storage is a Git service that other products embed. Customers expose clone, push and fetch endpoints inside their own applications, then use software development kits in TypeScript, Python or Go to create and manage repositories. Webhooks connect storage to build systems, bots and agents.
The published limits are aimed at machine workloads. Repositories can reach 32TB, and the service supports 30 or more concurrent writes. It also supports Git LFS and synchronizes with GitHub, GitLab and Bitbucket. Pierre lists unlimited request and API rate limits, with observed traffic of more than 500 Git requests per second per repository and 15,000 API requests per second.
The customer list is made up of AI-native builders. Pierre names Lovable, Poke, Anything, Bolt, Amp, Parahelp and Skydive. SiliconANGLE reports that Lovable cut code retrieval latency by about 70% using the service.
The broader bet is that repositories now hold more than source code. Pierre says customers store agent memory files, design artifacts and legal documents alongside code. That mix explains the pricing model, which separates hot storage from cold archives and charges for inbound and outbound bandwidth.
Pierre's team brings experience from large distributed systems. Its site says team members have worked at Cloudflare, Coinbase, Discord, GitHub, Meta, Reddit, Stripe and Twitter. That background fits a product that promises durable storage and consistent reads at high write rates.
Pricing is usage-based. Code Storage charges $0.005 per GB per hour for hot storage and $0.0002 per GB per hour for cold storage untouched for more than seven days. Bandwidth costs $0.06 per GB inbound and $0.15 per GB outbound. Plans start at $20 a month for multi-tenant use and $200 a month for a single-tenant option.
The architecture differs from Pando's. Code Storage says it uses Git's three-phase commit protocol and a quorum-based design to guarantee consistency across replicas. Pando accepts a looser consistency model in exchange for parallel writes. Bringing the two systems together would mean reconciling those choices, and OpenAI has not said how it will do that.
Is OpenAI trying to take on GitHub?
That is how several observers read the deal. SiliconANGLE reported that observers saw it as a shot at Microsoft-owned GitHub “for the age of artificial intelligence agents”. Axios framed its exclusive report as OpenAI targeting Microsoft's GitHub.
The comparison is easy to understand. GitHub is the central place where teams store code, review changes and coordinate releases. Codex is becoming a place where agents write, review and ship code. If Codex also stores the repository and its history, OpenAI controls more of the workflow.
OpenAI's language stops short of a replacement claim. The company says it wants to “advance the future of source control across the industry” and connect the code a company ships with the decisions its people make. It has not said it will remove Code Storage's existing GitHub sync.
The competitive pressure is real even if the products overlap only partly. GitHub is owned by Microsoft, and the deal takes OpenAI into a layer of the stack that Microsoft already controls. A move into source-control infrastructure puts OpenAI in a business adjacent to a major platform company.
Coexistence is part of the pitch. Code Storage already syncs with GitHub, GitLab and Bitbucket, so a team can keep its existing host while using Pierre's service underneath. OpenAI's acquisition does not remove that option, at least for now. The strategic question is whether Codex becomes the primary place where agent-generated code lives.
Developers should read the deal as infrastructure, not a new editor. The acquisition is about the pipes under Codex: repositories, history, review and storage. The model still writes the code, but the surrounding system decides how quickly that code can be reviewed, merged and deployed.
What changes for developers using Codex?
In the short term, very little. The deal has not closed, and the companies remain independent until it does. Trees and Diffs are already in the Codex app, and Pierre's open-source projects will remain open source.
After closing, OpenAI plans to bring Code Storage's features and functionality into Codex. That suggests repository creation, storage and history could become part of the Codex workflow rather than a separate service. OpenAI has not published a timeline or a product roadmap.
Existing Code Storage customers have a bridge. OpenAI says they can continue using the service after the acquisition closes while integration work proceeds. The company has not said whether the standalone product will remain on sale indefinitely, or whether prices will change.
For teams, the practical questions are about data and access. Will repositories move into Codex, or will Codex call out to Code Storage? Will permissions, audit logs and single sign-on carry over? Will the GitHub sync keep working for teams that treat GitHub as their system of record? None of those answers are public yet.
The migration path may be product-led. OpenAI's engineering post says its own engineers could ask Codex to migrate their development environments and services rather than moving them by hand. A similar approach inside Codex could connect repositories with limited manual work, though OpenAI has not described such a feature.
The open-source libraries are a separate matter. OpenAI says it will continue to contribute to Trees, Diffs and the other projects. Developers who depend on them should watch for changes in governance, release cadence and license text, even if the code stays open.
What are the limits of the deal?
The first limit is information. Without a price, it is impossible to judge whether OpenAI overpaid, underpaid or bought a team at a normal premium. The absence of headcount and retention details makes it harder still to assess what OpenAI is actually acquiring.
The second is technical fit. Pando and Code Storage solve overlapping problems with different consistency models. Pando allows readers of multiple branches to see different points in time. Code Storage promises consistent reads and writes across replicas through a quorum. Merging the products without breaking either promise is an engineering project, not a press release.
The third is evidence. OpenAI's performance numbers come from its own monorepo and production traffic. They show that Pando works at OpenAI's scale, but they do not prove that it will deliver the same gains for a smaller team or a different repository shape.
The fourth is regulatory risk. OpenAI says the deal needs approvals, and the companies will stay separate until they arrive. A delay would slow integration, and a challenge could end the deal.
The fifth is the open-source promise. OpenAI says Pierre's projects will remain open source and that it will keep contributing. That is a commitment, not a contractual guarantee visible to outside developers. The projects' health will depend on maintainers, reviewers and release processes after the team joins a larger company.
The sixth is adoption. Developers dislike changing the system that holds their history, permissions and CI hooks. Even a faster backend has to fit existing review rules, compliance audits and backup policies. OpenAI's internal success does not guarantee that outside teams will move.
What to watch next
Watch the closing. Regulatory approval is the first gate, and OpenAI has not given a date. Until the deal closes, Code Storage remains a Pierre product and the two companies remain separate.
Watch Microsoft's response. GitHub is the incumbent in code hosting, and Microsoft owns it. A public reaction, a pricing change or a new GitHub feature aimed at agent workloads would show how seriously the company takes the threat.
Watch the product roadmap. OpenAI has to decide whether Pando and Code Storage become one backend or stay separate. It also has to decide whether Code Storage remains a standalone commercial service or becomes an internal layer inside Codex.
Watch the developer experience. The deal is meant to make Git workflows faster for agents. Developers will judge it on push latency, merge reliability, review speed and whether the integration adds new lock-in.
Watch the open-source projects. Trees, Diffs and Pierre's other libraries will be judged by the developers who adopt them. Their release cadence and governance will show whether the open-source commitment holds after the acquisition.