openai acquires ona (formerly gitpod) — codex gets persistent cloud sandboxes for multi-day agent work

OpenAI announced on June 11, 2026, that it will acquire Ona — the Berlin-based cloud infrastructure startup that was, until September 2025, known as Gitpod. Terms were not disclosed. Once the deal closes (subject to customary regulatory approvals), Ona's entire team joins OpenAI's Codex group.

the practical outcome: Codex agents will be able to keep working for hours or days after you've closed your laptop, inside isolated cloud sandboxes, without your machine being present. that's a meaningful architectural shift — from “AI assistant you supervise” to “AI worker you hand a task and check on.”

what ona actually provides

Gitpod built its reputation on cloud development environments — disposable, reproducible workspaces spun up from a git repo in seconds. after the rebrand to Ona in September 2025, the company repositioned entirely around AI agent infrastructure. three components are at the core of what OpenAI is buying:

sandboxed cloud execution — isolated environments where an agent can run code, install dependencies, call APIs, and accumulate state across a long-running task without any of that leaking into the next agent run or into production systems.

workflow orchestration — tooling for sequencing and coordinating multi-step agent tasks, including retry logic, checkpointing, and handoff between sub-agents. an agent that gets interrupted mid-task can resume rather than restart.

enterprise security guardrails— access controls and audit logging built for environments where you can't have an agent with broad file system access just running loose. this is specifically the piece that enables Codex to operate inside enterprise clouds rather than only on OpenAI-controlled infrastructure.

the context: Codex at 5 million weekly users

OpenAI reported that Codex crossed five million weekly active users before this deal was announced. the sandboxed cloud execution gap is more obvious at that scale — developers running long refactors or multi-file agents on their local machines hit real constraints: sessions time out, laptops close, network drops. Ona's infrastructure directly addresses the bottleneck that shows up when you're not just prototyping with an AI assistant but actually delegating multi-hour work.

competitive read

Claude Code already runs headless via the SDK and can be pointed at long tasks in CI/CD environments, but that's an infrastructure-assembly exercise, not a first-party product. Cursor's background agent feature runs within the editor session. GitHub Copilot's agent mode operates within VS Code's lifetime. none of them have Ona's specific combination of “persistent cloud sandbox + enterprise-grade access controls + agent orchestration” as a shipping product.

the acquisition signals OpenAI's intent to own that layer — not build it from scratch — and to use it to create distance from competitors that are still shipping agent features bolted onto tools originally designed for single-session use.

honest take

the risk is integration. Gitpod/Ona was an independent product with an existing developer community and its own roadmap. acquisitions in the infrastructure layer tend to slow iteration in the short term while teams align on architecture. whether Ona's sandboxing technology becomes a first-class feature inside Codex within six months or eighteen is an open question. the announcement positions it as transformational; the execution will determine whether it actually is.

what's already real: Codex at five million weekly users is a product with serious adoption, and persistent cloud execution is a genuine gap versus what competitors currently offer. the deal makes strategic sense on paper. watch the shipping cadence post-close.