OpenAI targets long-running agents with acquisition of Ona

OpenAI has announced the acquisition of Ona, a startup that previously operated under the name Gitpod. Ona provides infrastructure that allows AI agents to run in cloud sandboxes rather than on developers' local machines. This enables tasks to be executed that take hours or even days, without requiring a developer to keep their own system available for that purpose.

The deal connects directly to Codex, the DevOps-focused tool OpenAI introduced earlier this year. With Ona's infrastructure, OpenAI can extend Codex to support more complex, long-running workflows that were beyond the reach of short-lived inference calls.

No financial terms of the acquisition have been disclosed.

From Gitpod to Ona: a deliberate change of direction

Gitpod was founded as a platform for cloud-based development environments, allowing programmers to work directly from the browser in a fully featured IDE. The company later repositioned itself as Ona, with a specific focus on running AI agents in isolated cloud environments.

That pivot proved strategically well-timed. Where Gitpod competed in a market alongside players such as GitHub Codespaces and Coder, Ona targets an emerging need: providing stable, scalable execution environments for autonomous AI systems that independently write, test, and modify code.

Cloud sandboxes are essential in this context. They offer isolation, reproducibility, and the ability to run multiple agents in parallel without interference, properties that local development environments structurally lack when handling workloads of any significant scale.

Codex gains an execution foundation

Codex, which OpenAI announced in May 2025, is a cloud-based software engineering agent capable of independently performing tasks such as writing functions, resolving bugs, and navigating codebases. The agent operates asynchronously, meaning a user can submit a task and retrieve the result later.

Until now, the execution time for such tasks has been limited. With Ona's infrastructure, OpenAI aims to raise that ceiling. Agents should eventually be able to handle tasks spanning multiple hours or days, for example, refactoring large codebases, running extensive test suites, or working through iterative build-and-validation cycles.

That requires more than a powerful language model. It demands a reliable execution layer that manages processes, handles errors, and tracks the state of a task over an extended period of time. That is precisely what Ona built.

Broader trend toward autonomous software development

The acquisition fits into a wider movement in which AI companies are not only building models, but also putting in place the infrastructure to let those models operate autonomously. Anthropic is working on so-called computer use capabilities, Google DeepMind is investing in agentic systems through Gemini, and startups such as Cognition, maker of Devin, are focused entirely on autonomous software engineers.

With this acquisition, OpenAI is opting for vertical integration: the company will control both the model and the environment in which that model executes tasks. This reduces dependence on external cloud providers for the execution layer and provides greater control over performance, security, and costs.

For developers using Codex, this means in theory that they can delegate more complex tasks without having to think about the underlying infrastructure themselves. The sandbox environment handles isolation and reproducibility, while the model carries out the task itself.

Questions around integration and availability

It remains unclear when Ona's technology will be fully integrated into Codex and what implications this will have for existing Ona customers and partners. OpenAI has not published a timeline for the rollout of long-running agent functionality.

There is also the question of how OpenAI will handle the enterprise branch of Gitpod, which was still active at the time of the rebranding to Ona. Companies that depend on that service will want to know whether their environments will be continued or phased out.

OpenAI has not yet provided a public answer to these questions. The company announced the deal via a brief statement, without further technical details on the integration. Additional information is expected to follow with future product announcements around Codex.