OpenClaw 2.0 Turns Personal Agents Into a Shared Platform
The open-source agent adds remote sessions, interactive dashboards and scoped secret handling in a major architectural overhaul.
From local assistant to distributed workspace
OpenClaw has released version 2026.8.1, branded OpenClaw 2.0, with a rebuilt web application and a broader architecture for running agent sessions across personal devices and cloud workers. Release artifacts, including signed and notarized macOS packages, became publicly downloadable on August 31.
The update allows a session to move beyond its original Gateway—the component connecting an OpenClaw installation to tools and messaging services—and continue on paired devices or remote workers. Users can preserve warm machines and seeded project environments for later tasks. A durable progress card follows long-running work through page reloads while exposing subagent activity and accumulating file changes.
OpenClaw 2.0 also introduces interactive widgets and dashboards inside conversations. Agents can present structured questions, request approval or display generated interfaces, while users can pin results and restrict which actions or network origins a widget may access. Conversation search can reopen surrounding messages from an exact text match.
Security receives several targeted changes. An agent can request credentials through a masked interface that keeps the secret out of both chat history and model context. An optional proxy substitutes protected values only for approved destinations. Users can also approve a precisely defined recurring operation once, inspect or revoke that permission later, and force renewed approval when the operation changes.
Why it matters
The release reflects a shift in open-source agents from single-machine chat wrappers toward persistent operating environments shared by people, devices and subordinate agents. Features such as session mobility and reusable remote workers make OpenClaw more practical for sustained projects, but they also enlarge the boundary within which permissions and credentials must remain trustworthy.
Its most consequential work is therefore less visible than the redesigned interface: credentials, recurring approvals and interactive outputs now have explicit control surfaces. Whether those controls withstand malicious webpages, compromised plugins and ambiguous user instructions will determine if OpenClaw 2.0 can support serious automation rather than merely more elaborate demonstrations.