OpenAI Acquires Ona to Give Codex Agents a Persistent Cloud Desk
OpenAI has agreed to acquire Ona, a startup that builds secure, persistent cloud execution environments for AI agents. Ona will join the Codex team after closing.
The stated rationale is direct: Codex is outgrowing the browser tab. At 5 million weekly users — up 400% in recent months — the bottleneck is no longer model intelligence. It is what happens when an agent needs to run for six hours, access live credentials, log into systems, and pick up where it left off after the human disconnects.
What Ona Adds
Ona provides customer-controlled cloud workspaces. An agent running inside Ona has access to the full stack a real engineer uses: a persistent filesystem, build tooling, runtime environments, credentials scoped to the task, execution logs, and audit trails. The workspace stays alive after the laptop closes. The agent keeps working.
That runtime persistence is the gap Ona fills. Current Codex deployments are session-bound. Every new interaction starts from a clean slate. For short tasks — a function stub, a test suite, a bug fix with a clear repro — that works. For a multi-day database migration, a security audit spanning 40 services, or a refactor that requires running the test suite 200 times, session-bound execution breaks the workflow.
The Enterprise Unlock
The harder commercial argument is about where agents are allowed to operate.
Most large enterprises are unwilling to route sensitive code, internal credentials, or regulated data through an external sandbox they do not control. A bank, a pharma company, or a sovereign wealth fund will not give a general-purpose cloud agent standing access to production systems. The review trails do not exist. The access controls do not fit their compliance requirements.
Ona solves this by running inside the customer’s own cloud boundary. The agent operates within the customer’s VPC, under the customer’s IAM policies, logging to the customer’s audit infrastructure. OpenAI provides the model and the orchestration layer. The enterprise provides the compute and retains control of what the agent can see.
That architectural split unlocks the enterprise tier that Codex cannot currently address. Longer-running work categories — legacy codebase migrations, multi-system incident investigations, vulnerability remediation at scale, test coverage generation across large repos — all require this model.
What Changes for OpenAI
The acquisition is a bet that the economics of AI coding shift toward durable workflows.
Codex’s current billing is session-based and token-bound. An enterprise paying for Codex today pays for the conversation. Persistent execution environments let OpenAI charge for the workflow: a multi-day migration project, a monthly security scanning service, an ongoing refactoring engagement. These categories have materially higher contract values than per-session token consumption.
OpenAI has been building toward this. The June 2026 redesign of ChatGPT around agents — internally described as “chat is dead” — is the platform context. Ona is the infrastructure layer that makes enterprise agent work viable.
OpenAI said Ona’s technology will help Codex take on longer-running work and help more organizations deploy agents securely in production. No acquisition price was disclosed.