2026-06-17 — views

OpenAI Acquires Ona (Formerly Gitpod) to Give Codex Agents Persistent, Secure Cloud Environments

Read this because The race in agentic coding has shifted from model quality to runtime. OpenAI is buying the boring, hard infrastructure — persistence, sandboxing, audit — that turns a clever agent into a deployable one.

OpenAI is acquiring Ona — the cloud dev-environment firm formerly known as Gitpod — to let Codex agents run longer tasks in persistent, audited environments.

What is happening

OpenAI announced on June 11, 2026, that it has agreed to acquire Ona — the cloud development-environment company formerly known as Gitpod — to bring secure, persistent cloud execution into its Codex coding-agent ecosystem. Ona is a roughly 79-person team that provides cloud development environments (CDEs): isolated, persistent workspaces where code can be built and run. OpenAI’s stated rationale is that Ona “provides secure, persistent environments where agents can access the tools, systems, and context they need to make progress over time.”

No purchase price was officially disclosed. IDC analyst Arnal Dayaratna estimated the deal at roughly $450–500 million against assumed 2026 revenue of $10–15 million.

Why OpenAI bought a dev-environment company

The constraint on coding agents is no longer just how smart the model is — it is where the agent runs and for how long. A capable model still fails in production if it is trapped on a single device or limited to one active session. Ona’s technology is aimed squarely at that gap.

Capability	What it enables for Codex agents
Persistent environments	State survives across multiple sessions, not just one chat
Customer-controlled infra	Agents run inside the enterprise’s own security perimeter
Secure tool/system access	Credential management, governed access to internal systems
Audit and governance	Audit trails for what an agent did and why

The shorthand: OpenAI is buying the runtime, not the model. Codex already has frontier-grade reasoning; what it has lacked is a place to act over hours or days, inside infrastructure a company actually controls, with the audit trail security teams require before they will let an agent touch production.

The strategic read

This is the agentic-coding race playing out in infrastructure rather than benchmarks. Devin, Claude Code, and Codex have converged on a similar premise — agents that work on real repositories over long horizons — and the differentiator is increasingly the execution substrate: sandboxing, persistence, secrets handling, and observability. Acquiring Gitpod’s successor gives OpenAI a mature CDE built precisely for that, rather than building it in-house from scratch.

It also fits a broader June pattern for OpenAI around Codex, including expanded availability of frontier models and Codex on AWS earlier in the month. The Ona deal is the infrastructure complement: distribution plus a hardened place to run.

Why this matters for builders

For teams evaluating coding agents, the practical lesson is that runtime is now a first-class selection criterion. Ask not only how a model scores on SWE-bench, but where its agents execute, whether state persists between sessions, whether the environment sits inside your security boundary, and what audit trail you get. Those are the properties that decide whether an agent clears an enterprise security review — and they are exactly what OpenAI just paid to own.

If you currently run Gitpod/Ona environments, watch for integration timelines and any changes to standalone availability or pricing as the team folds into OpenAI; acquisitions of developer tooling frequently reshape the open or self-hostable tiers. If you are betting on a coding-agent platform, weigh the execution layer as heavily as the model — a slightly weaker model in a persistent, audited, customer-controlled environment may be more deployable than a stronger one confined to an ephemeral session.

The bottom line: the agentic-coding contest has moved past “whose model is smartest” into “whose agents can safely run long enough to finish the job.” OpenAI just bought a direct answer to the second question.