OpenAI adds Ona-backed persistent runtimes to Codex

• OpenAI acquisition post says OpenAI will acquire Ona to bring secure cloud execution and orchestration into Codex, aimed at agent jobs that keep running for hours or days instead of a single laptop session.
• OpenAI also said in the acquisition post that Codex has passed 5 million weekly users, up 400% from earlier this year, which helps explain why background execution has become product infrastructure instead of a side feature.
• testingcatalog's summary and OpenAI's announcement both frame the deal around persistent runtimes: Codex work should keep progressing after the original device is closed.
• Ona's statement and Ona's product page show what OpenAI is actually buying: customer-controlled cloud environments, background agents, scoped credentials, audit trails, and orchestration for repeatable jobs.

You can read OpenAI's announcement, compare it with Ona's own note, and then open the Codex cloud docs and cloud environments docs to see the current model: GitHub-connected background tasks, containerized runs, and a cache that lasts up to 12 hours. Ona's product page fills in the missing infrastructure picture, background agents that run in full cloud environments with network access, scoped credentials, audit trails, and VPC deployment.

The core change is not a new model. It is where Codex jobs live.

According to OpenAI's announcement, Codex's most valuable work is increasingly unfolding over hours or days, and OpenAI wants that work to continue outside the machine and session where it started. The post says Ona gives Codex secure, persistent environments where agents can keep access to tools, systems, and context over time.

That lines up with testingcatalog's summary, which reduced the whole deal to a simple product shift: secure cloud execution so Codex can keep working even when the laptop is closed.

Ona's own materials make the acquisition feel less speculative than the announcement headline. On its product page, Ona describes itself as a platform for background agents with three concrete pieces:

• Background agents that turn a task into a pull request.
• Full cloud environments with tools, network access, and permissions.
• Governance features such as VPC deployment, audit trails, scoped credentials, and kernel-level policy enforcement.

In Ona's statement, the company adds a few more enterprise details that OpenAI's post only hints at: reproducible environments, repeatable automations, runtime AI security, and work that carries state across tools and sessions. Ona also said weekly agent sessions had grown 13x this year across large customers in banking, pharma, and finance.

OpenAI already has a cloud version of Codex, so the interesting part is the gap between today's docs and the infrastructure Ona advertises.

The Codex web docs say Codex can already run tasks in the background, including in parallel, inside OpenAI-managed cloud environments tied to GitHub repos and pull requests. The cloud environments docs get more specific:

• Each task starts in a container with the target repo checked out.
• Setup scripts can install dependencies and tools.
• Agent internet access is off by default, but can be opened up.
• Container state is cached for up to 12 hours.
• Secrets are encrypted and removed before the agent phase.

That is already a real remote execution system. The Ona deal suggests OpenAI wants a heavier-duty version, one built around persistent enterprise-controlled runtimes instead of shorter-lived cloud task containers.

The acquisition is not closed yet. OpenAI's announcement says the deal is still subject to customary closing conditions and regulatory approvals, and that OpenAI and Ona remain separate companies until then.

That same post says the Ona team will join Codex after closing. Ona's statement says existing customer commitments stay in place in the meantime, which matters because this is not a talent acquihire tucked into a lab org. It is a live product being pulled directly into Codex.