OpenAI has agreed to acquire Ona, a German startup that retains software program brokers working in a safe cloud after the developer who began the duty logs off. The deal was introduced Thursday and folds Ona into Codex, OpenAI’s coding agent. It factors the corporate at an issue that higher fashions alone haven’t solved.
Codex now reaches greater than 5 million folks every week, OpenAI stated, a 400% leap for the reason that begin of the yr. The work it does has stretched from minutes into hours and typically days. Jobs that lengthy want an execution atmosphere to run. For many enterprises the tougher query is a governance one. Will a financial institution or a hospital let an autonomous agent function inside its personal community, attain its knowledge, and preserve working whereas no one is watching? Ona is OpenAI’s reply to that query.
Inside Ona
To understand what Ona provides, it helps to know the place it got here from. The corporate started as Gitpod, a Kiel-based developer-tools agency that moved coding off native machines and into the browser. It says it has served 2 million builders. In late 2025 it rebranded to Ona and rebuilt itself round brokers.
The Ona platform is constructed from three constructing blocks. The primary is environments: sandboxed cloud workspaces outlined in code that spin up the identical means each time. The second is brokers, background staff that settle for a process and return a pull request, reachable from any gadget as soon as the work is operating. The third is a set of guardrails, the controls a safety workforce cares about. They embody audit trails, role-based entry, scoped credentials and deployment inside an organization’s personal digital personal cloud.
What ties these collectively is an thought Ona calls customer-controlled execution. OpenAI provides the mannequin and the orchestration. The agent runs contained in the buyer’s cloud, the place the corporate retains its knowledge, its credentials and its audit path. With this, an organization can let Codex modernize a legacy codebase or patch a category of vulnerabilities throughout days of labor. The agent operates on infrastructure the corporate controls moderately than on OpenAI’s servers.
Why OpenAI Wants Ona
Codex started as a instrument for software program builders, and OpenAI has pushed it properly previous that base. Data staff now account for roughly one in 5 Codex customers and are rising quicker than the core developer group, the corporate stated. Plugins have prolonged it into gross sales, funding banking and fairness investing. As extra of that work runs unattended over lengthy stretches, the place it executes stops being a element and turns into the factor an enterprise purchaser has to log out on.
That’s the place Ona earns its preserve. Its current prospects embody banks, pharmaceutical corporations and sovereign wealth funds. These are organizations that care much less a few leaderboard rating and extra about the place the code runs and who can see it. By bringing that execution layer in-house, OpenAI can inform a regulated purchaser {that a} Codex agent will function underneath the customer’s personal safety and governance guidelines.
The Aggressive Image
The acquisition is a belief pitch aimed straight at cautious IT departments, and the timing isn’t any accident. OpenAI is racing Anthropic, whose Claude Code has grown rapidly inside engineering groups over the previous yr. Each corporations have filed confidential prospectuses with the SEC. OpenAI carries a valuation round $852 billion, so every is underneath stress to show coding-agent enthusiasm into manufacturing income.
Ona suits a sample in OpenAI’s latest shopping for. The corporate picked up the security-testing startup Promptfoo in March and the healthcare-technology agency Torch in January. Each offers added enterprise plumbing moderately than uncooked mannequin functionality.
The Limits
For all that, the deal doesn’t shut the case. It’s nonetheless topic to regulatory approval. A US large absorbing a German firm that builds agent infrastructure may draw scrutiny on each side of the Atlantic earlier than the Ona workforce formally joins OpenAI.
A sharper query hangs over mannequin neutrality. As an impartial vendor, Ona permits enterprises to attach their very own fashions by way of providers akin to Amazon Bedrock and Google Vertex AI. A buyer may run Claude or Gemini inside an Ona atmosphere. Beneath OpenAI’s possession, that openness is value watching, and any purchaser that selected Ona for multi-model flexibility has motive to ask how lengthy it’s going to final.
The deeper restrict sits under all of this. Buyer-controlled execution settles the place an agent runs, not whether or not the agent stays right over a two-day job. An autonomous agent that grinds away for 2 days can be fallacious for 2 days. The instruments for reviewing lengthy unattended runs are far much less mature than the environments that host them. Working these brokers inside your personal digital personal cloud additionally strikes actual operational work onto your platform workforce, who now personal the infrastructure the brokers stay on.
The Street Forward
For a CIO weighing this, the questions are concrete. Ask whether or not Ona will keep model-neutral underneath OpenAI or quietly favor Codex. Press on which guardrails are genuinely Ona’s personal and that are skinny wrappers round AWS or Azure controls you already pay for. And learn how evaluation and rollback work when an agent has been operating for a day inside manufacturing programs.
The Ona acquisition exhibits that the competition amongst coding brokers is shifting from the mannequin to the execution layer. That layer decides whether or not a regulated enterprise will let an agent contact its programs in any respect. OpenAI is shopping for its means into that layer moderately than constructing it. That may be a calculated transfer, given how far forward Ona already sits on enterprise controls. Loads of corporations needed the productiveness of autonomous brokers however balked at handing their code to a vendor’s cloud. For them, customer-controlled execution opens a reputable path to manufacturing on their very own phrases.
