OpenClaw 2026.4.5 Update: Native Video Generation + 'Dreaming' Memory System—Thriving Despite Anthropic Ban

When Anthropic cut off OpenClaw’s free Claude access last month, I thought the project was done. Instead, they dropped a major release on April 5th that basically said: ‘If you won’t let us use your API, we’ll build our own.’

Native Video Generation: Not Just API Wrapping

The headline feature is native video generation. This isn’t a thin wrapper around third-party APIs—OpenClaw integrated 11 different video providers directly. One command, and the AI handles the rest.

I tested it. From prompt to finished video, the workflow is smoother than running Kling or Pika separately. The kicker? The AI can automatically pick which model to use based on your scene.

This is what MCP was supposed to enable: AI agents choosing their own tools instead of being hard-coded to specific vendors.

The Dreaming System: AI That ‘Sleeps’

Sounds mystical, but the logic is solid. OpenClaw added a /dreaming command that simulates how humans consolidate memories during sleep.

When inactive, the AI actively organizes previous conversations and knowledge into what’s called a ‘Memory Palace.’ Ask about something you discussed days later, and it actually remembers.

I ran it for a week. The difference is real. A technical solution we talked about three days ago? The agent still referenced our earlier conclusions accurately. Standard agents can’t do this.

Simplified Chinese UI: Finally

Previous versions were English-only, which was annoying for Chinese users. This update adds proper Simplified Chinese support—and the translation doesn’t suck.

Getting Banned Made the Product Better

Here’s the irony. Anthropic cutting off free access should have been a death blow. The community response? ‘Fine, we’ll support every model instead.’

Now OpenClaw doesn’t depend on any single provider. GPT, Claude, Gemini, domestic models—all supported, with automatic switching based on task type. This decentralization actually makes it more robust.

I’m increasingly convinced that the future of AI tools isn’t vendor lock-in. It’s platforms like OpenClaw that use protocols and ecosystems to break monopolies.

There are rough edges. Dreaming hogs more resources than expected. Setting up 11 different video API keys is still a pain. But the trajectory is clear.

If you’re looking for a truly comprehensive AI Agent platform, OpenClaw’s latest release deserves serious consideration.