OpenAI Group PBC reveals its intention to purchase Ona, a startup offering a platform for managing enduring artificial intelligence agents.

Developers frequently execute the AI agents they utilize for coding on their local computers. When a workstation is powered down, the AI agents it operates cease their tasks. This restricts their capacity to undertake extended tasks that last several days.

Ona, known officially as Gitpod GmbH, offers a platform that allows AI agents to operate in cloud-hosted sandboxes. Ona sandboxes stay active even when the developers using them turn off their workstations, ensuring that AI agents’ tasks continue uninterrupted.

The firm claims that its software also offers additional advantages. The cloud provides greater computing resources than a developer workstation, allowing agents to finish tasks more rapidly. Moreover, Ona removes the sandboxes where it operates AI agents when they are no longer required. Local AI agent environments may not be removed quickly, raising the likelihood of data theft.

Administrators have the ability to set up an Ona sandbox to block agents from reaching potentially harmful applications. The company states that its platform is capable of blocking a harmful application even if hackers change its name, relocate it to another file path, or conceal it within a script.

The software identifies obfuscation efforts through a technique known as hashing. It creates a hash, or distinct cryptographic identifier, for each program that administrators intend to restrict. The signature allows Ona to recognize a blocked application even when its file name or other parameters are altered.
The platform of the company also features additional security measures.

Also Read: NVIDIA's Rubin Moment at CES: AI Supercomputing Breakthrough

It can inhibit AI agents from reaching areas of the file system that hold sensitive information like encryption keys. Ona states that the software additionally prevents outgoing connections to potentially harmful servers.

OpenAI plans to utilize the firm's technology to improve its Codex AI assistant. Codex was introduced last year with an emphasis on accelerating software development initiatives. This month, OpenAI introduced enhancements that improve the tool's effectiveness in non-technical activities like generating ad graphics. The firm reports that Codex attracts over 5 million users each week.

Also Read: AI Autonomy Race: How Advanced are Top Countries' AI Strategies?

OpenAI states that Ona will enhance Codex’s capability to handle long-duration tasks that require hours or even days. Specifically, it will allow users to evaluate agents' work and offer feedback when needed. The Ona team will collaborate with the Codex team to assist OpenAI in executing the enhancements.

Also Read: From Silos to Synergy: Building the Future With Agentic AI

The agreement follows three months after the ChatGPT creator's last purchase. In March, OpenAI acquired Promptfoo Inc. to improve the cybersecurity capabilities of its OpenAI Frontier platform, allowing businesses to develop tailored AI agents. Promptfoo created a collection of tools capable of identifying and resolving weaknesses in AI applications.