OpenClaw has shipped its largest update since launch. The project calls the release OpenClaw 2.0. Nine hundred thirty-three contributors built it. More than 16,000 pull requests went into the code. Meanwhile, 569 of those contributors submitted work for the first time.
The team spent nearly seven weeks on the release. Earlier, OpenClaw had shipped 106 updates in 230 days. Most of those arrived within one or two days of each other. In contrast, this cycle took much longer because the work spread across the whole platform.
The update changes installation, messaging, memory, skills, models, and automations. It also updates the browser app, native apps, plugins, and security. According to the official blog, the team first aimed to simplify setup and rebuild the browser. After that, the cleanup spread through the rest of the product.
New users can now start with tools already on their computers. Existing ChatGPT or Claude subscriptions work. API keys work as well. Local models also work. As a result, people can reach a first conversation sooner. Later, they can finish remaining setup by talking to the agent.
OpenClaw rebuilt the browser app as a main workspace. Users can continue setup there. They can also return to ongoing tasks. In addition, they can watch work happen in real time.
Users can search past conversations by exact words or phrases. A match then opens the nearby messages. Sessions can also run beyond the local Gateway. Work can move to paired devices or cloud workers. The session workspace travels with it. Warm machines and project seeds can be reused later.
Durable progress cards stay visible after a reload. They show subagent activity and growing edits. Widgets and dashboards add more interaction. Users can place widgets inside chats. They can pin them to session dashboards. They can also grant limited actions or network origins and export views as images.
For repeated tasks, a user can approve one operation once. After that, an automation can run the same step again. Media support is wider now. Native playback controls are new as well.
Security is tighter too. An agent can request credentials through a masked prompt. The secret stays hidden from the chat and from the model context.
Even so, OpenClaw remains self-hosted and open source. Developers can keep building on it without locking into one AI model or one provider.