[DigitalToday reporter Hwang Chi-gyu (황치규)] OpenClo 2.0, an open-source AI agent tool, has been released. Version 2.0 is the first major update since OpenAI hired OpenClo developer Peter Steinberger.

Unlike a general chatbot that answers questions within a browser tab, OpenClo is an open-source personal AI agent framework that carries out real tasks on a user’s behalf. It runs on a personal PC or in the cloud and has been used for inbox cleanup, schedule management, checking daily activities and coding assistance.

On Aug. 31, local time, SiliconANGLE reported that OpenClo 2.0 is the largest update in OpenClo’s history. It focuses on usability improvements, stronger state persistence, simplified installation and enhanced security.

The update simplifies installation. Users can automatically detect existing ChatGPT or Claude subscriptions, API credentials and locally hosted models on a PC or cloud instance. Some settings apply by default, allowing users to start talking with the agent more quickly.

The dashboard has been redesigned as a unified control screen. With the chat window placed at the front, users can start new conversations or return to tasks in progress and watch the AI agent work in real time. New widgets and dashboard features also let users bring widgets into the chat window or pin them to the dashboard, and specify actions that run based on events.

Collaboration features have also expanded. Previously, multiple users could not work together with the same AI agent, but shared cloud sessions now allow teammates to join an ongoing session and continue the work. It has also become possible to hand off tasks to colleagues while preserving the full context, or to carry out content work and automation together in real time. The development team is also using the multiplayer feature for OpenClo framework code development.

Search has been improved so users can find exact wording, words or phrases in past conversations and reopen surrounding messages. On the security side, explicit trust boundaries, least-privilege controls and protections against prompt abuse have been introduced. Permissions are separated based on who can run the agent, what it can access, where code is executed and where credentials can be sent.

Because early versions and installation environments had major security vulnerabilities, the overhaul also focuses on reducing the attack surface. Even so, SiliconANGLE reported that if users connect personal information or use tools that can run in cloud services or on local devices, they should protect the installation environment from the public internet and maintain strong authentication.