OpenAI Acquires Ona to Enable Persistent, Long-Running Codex Agents

OpenAI has made a significant strategic move by acquiring Ona, a cloud startup renowned for its secure and persistent execution environments tailored for AI agents. This acquisition, announced on June 11, 2026, aims to bolster OpenAI's Codex product line, enabling its AI agents to handle more complex and extended tasks. Previously known as Gitpod before a 2025 rebranding, Ona has pivoted from browser-based development environments to focusing on autonomous software-engineering agents and isolated workspaces. The core benefit of integrating Ona's technology is the ability for Codex agents to maintain state and continue working on tasks that span hours or even days, such as large-scale codebase refactoring or comprehensive test suite executions, even when a developer disconnects. This addresses a long-standing limitation of AI coding tools, which have often been constrained by the session length and local hardware capabilities of developers. Ona's sandboxed environments provide the necessary infrastructure, including source code, dependencies, tools, and scoped credentials, to ensure secure and isolated execution within customer-controlled cloud accounts. For enterprises, this acquisition is particularly impactful as it offers enhanced governance and control over AI agent activities. IT teams will gain the ability to dictate where code executes, which resources are accessible, and how activity is logged and reviewed, mitigating common concerns about data access and security. Industry analysts suggest that this focus on long-running agentic infrastructure represents a new competitive frontier for OpenAI, especially against rivals like Anthropic, which may now face pressure to develop similar persistent sandbox capabilities. The entire Ona team, including its CEO Johannes Landgraf, is set to join OpenAI's Codex group, pending regulatory approvals. This integration is expected to accelerate the development and deployment of enterprise-ready AI agents capable of performing sophisticated, long-horizon work. The move underscores the evolving landscape of AI development, where the emphasis is shifting beyond just model improvements to include robust lifecycle management, memory, and interface controls for production-grade agents.