OpenAI plans Pierre acquisition to bring Code Storage into Codex

OpenAI announced plans to acquire developer-tools company Pierre on November 4 and bring its Git-based Code Storage features into Codex. Codex already uses Pierre libraries.

Closing requires regulatory approvals and other customary conditions. The companies remain independent until then. OpenAI says Code Storage customers can continue using the service after closing, and Pierre’s open-source projects will remain open-source.

Code Storage holds generated files and their history using Git. Pierre describes uses beyond source code, including agent memory, design artifacts and legal documents that change over time. The company names Lovable and Bolt among services using its infrastructure.

Applications can expose Git clone, push and fetch endpoints inside their own products while accessing files and history through programming interfaces. This supports AI-product builders creating and updating repositories programmatically.

Agent attempts stay separate through review

Code Storage’s parallel-attempts guide starts multiple temporary branches from the same code revision. Creating branches writes references without copying the repository’s Git objects. Each agent records a separate attempt. The application compares every result against the common starting point and evaluates it through tests, code review or another assessment. A selected attempt becomes a normal branch for review and merging.

The service can also mirror existing GitHub repositories. In GitHub App mode, GitHub remains the source of truth: Code Storage forwards pushes there and copies changes back. Normal branches and tags synchronize; temporary branches stay outside that mirror. A separate credential-free mode imports a public repository once. Continuous two-way synchronization uses GitHub App mode.

Pierre also builds the interface for people examining changes. Its Diffs library displays old and new files side by side or in a unified view, highlights changed words or characters, and marks added and removed lines. Applications can attach review comments, automated-check results and custom accept-or-reject controls.

The separate editing feature lets reviewers change code directly in a comparison, search and replace, and undo edits. Demonstrations also show collaborators’ cursors and selections. Pierre labels this editing mode experimental. These documented library capabilities do not establish that every feature newly shipped in Codex with the acquisition announcement.

For agent memory, the storage guide pins reads to a particular revision and accepts a write only while the target branch still matches that starting revision. If another agent saved first, the application reloads current files and applies its change to that version. This guards against overwriting newer work with an older file. Separate memory repositories can carry different access rules or retention policies for projects and users. Restoring earlier content creates a new commit, preserving the undo action in history.

Code Storage’s existing pricing starts at $20 in minimum monthly usage for Starter and $200 for Scale. Those amounts are minimum spending commitments, not fees added on top of usage. They describe the storage service, rather than a newly announced Codex subscription price.

Hot storage costs $0.005 per gigabyte-hour, while cold storage untouched for more than seven days costs $0.0002. Pierre positions hot storage for low-latency access and cold storage for compressed, inactive code. Enterprise plans offer dedicated environments and negotiated usage commitments.