OpenAI Acquires Ona to Give Codex a Persistent Cloud Brain for Multi-Day AI Tasks
OpenAI is acquiring Ona, the secure cloud execution startup, to let Codex run long-horizon agent tasks inside customer-controlled environments.
OpenAI has announced it will acquire Ona, a startup that builds secure, cloud-based execution environments for AI agents. The deal is subject to standard regulatory approvals, and financial terms have not been disclosed. Once it closes, Ona’s team and technology will move into OpenAI’s Codex division.
The acquisition is a fairly direct answer to a specific problem: Codex has outgrown the session model.
The Problem Codex Is Outgrowing
Right now, most AI coding tools work the way a browser tab works. You open a session, give the model a task, and the work happens while you’re there. Close the laptop, end the session. That’s fine for quick tasks, but it becomes a real constraint as the tasks get more ambitious.
OpenAI’s announcement frames it plainly: as Codex becomes more capable, its most valuable work unfolds over hours or days. The goal is for people to hand off a meaningful piece of work and trust it will keep running, with Codex reporting back progress, flagging decisions, and completing results from wherever you happen to be next.
That requires persistent, secure compute running somewhere other than your local machine. That’s what Ona provides.
What Ona Actually Does
Ona was founded in 2019 as Gitpod, one of the better-known cloud development environment tools, before rebranding in September 2025 to focus specifically on AI agent infrastructure. The rebrand reflected a deliberate pivot: CEO Johannes Landgraf’s view was that IDEs defined the last era of software, and agents define the next.
The product gives AI agents a sandboxed, ephemeral workspace in the cloud, complete with short-lived credentials, audit logs, a command deny list, and controlled access to external resources. It runs either in Ona’s own multi-tenant cloud (with US and EU regions) or inside a customer’s AWS VPC for tighter isolation. Ona currently supports more than 2 million developers and already shares customers with OpenAI. Enterprise adoption of its agent capabilities has grown 13-fold in 2026 alone.
Its clients include a major US bank, European pharmaceutical companies, and Asian sovereign wealth funds. These are not organisations that will let an AI agent anywhere near production systems without knowing exactly where it’s running, what it can touch, and who reviewed what.
What This Means for Enterprises
This is where the acquisition matters most, at least in the near term.
OpenAI’s Core Products Lead Thibault Sottiaux put it directly: “Enterprises want powerful agents that can do real work while meeting the security and control requirements of their environments. Ona will help us make Codex easier to deploy securely across production workflows for customers operating at the highest standards of trust and scale.”
The practical implication is that organisations will be able to run Codex agents inside their own cloud environment, rather than sending sensitive workloads into OpenAI-controlled infrastructure. That means control over where agents run, what they can access, how credentials are scoped, and how activity is logged. For regulated industries, that level of governance is often a prerequisite, not a preference.
The use cases OpenAI is targeting include software testing, application modernisation, vulnerability management, and issue resolution. These are all tasks that genuinely benefit from continuous execution over time, not just a single assistant response.
What This Means for Developers
For individual developers, the immediate impact is limited. The acquisition hasn’t closed yet, and Ona’s infrastructure is not yet integrated into Codex. But the direction is clear enough.
Codex is being built toward something closer to a persistent agent than a session-based assistant. The idea is that you hand it a multi-hour task, it runs in the cloud, and you check back in from whatever device you’re on when it’s done. That’s a meaningfully different product from what most people are using today.
Codex now has more than 5 million weekly active users, up roughly 400% from earlier this year. Perhaps more telling: knowledge workers now make up around 20% of that user base and are growing at triple the rate of developers. The product’s centre of gravity is shifting, and persistent cloud execution broadens what it can realistically be used for.
A Pattern of Targeted Acquisitions
This is part of a consistent approach from OpenAI. Earlier this year it acquired Torch, a healthcare tech startup, for roughly $60 million, and Promptfoo, a cybersecurity startup focused on LLM vulnerability testing. Each acquisition has addressed a specific capability gap rather than general expansion.
Ona fits the same pattern. OpenAI is not trying to build cloud infrastructure from scratch. It is acquiring a team that has already solved the hard problems of sandboxed, auditable, enterprise-grade agent execution and folding that expertise into Codex.
The competitive context is real too. Anthropic’s Claude Code is widely considered the current leader for long-running coding tasks. Persistent, customer-controlled execution environments are a meaningful piece of closing that gap.
The Takeaway
If you are an enterprise evaluating Codex for anything beyond one-shot tasks, this acquisition is worth watching. The combination of OpenAI’s model capability and Ona’s execution infrastructure is pointed squarely at the governance and continuity requirements that have kept many organisations from deploying AI agents in production.
If you are a developer using Codex day-to-day, the near-term change is modest. But the longer arc is toward a tool that you can genuinely delegate to, trust to keep working, and check back in with rather than babysit. That is a more useful product, and this acquisition is a concrete step toward it.