This was a talk I gave at the OpenClaw Technical Workshop at the 2026 China Generative AI Conference (Beijing). Officially, I was there as VP of Qujing Technology, but I preferred to speak from the perspective of someone who spends every day helping people install OpenClaw.
Why This Talk
After helping dozens of people install OpenClaw remotely, one thing became increasingly clear to me: the OpenClaw agent paradigm has pushed the question of "hardware substrate" back to center stage. In the past, a computer was just a computer. Now you have to decide upfront: which machine does the agent run on? Does it stay on 24/7 or require human oversight? Is the data local or in the cloud? Who is liable when something goes wrong?
None of these questions can be answered cleanly in a purely edge or purely cloud paradigm. Edge is cheap, private, and capable of asynchronous computation, but its intelligence hits a ceiling early. The cloud is plenty smart, but privacy, cost, and legal liability all become blockers. For the foreseeable future, the only viable path is edge-cloud integration. This talk is about that path.
Three Key Arguments
- Edge devices should stop chasing large models. Compute, bandwidth, and memory supply chains are three mountains pressing down at once. What the box should pursue isn't "can run 70B," but rather "stays awake, stays cheap, stays quiet."
- Multimodal is what the edge is actually worth running. Embedding, ASR, TTS, OCR, VLM—these small models are used daily. What they have in common is that you don't want them leaving the device, and they aren't time-sensitive. Meanwhile, the cloud charges by the second at absurd rates.
- AIMA makes edge-cloud integration a walkable path. From installation to connection to interaction, it ties everything together in one go. You tell the cloud, "Install OpenClaw on this machine, connect LLM, connect Feishu," and the edge and cloud start moving together immediately.
Why "More Secure"
I spent a good portion of this talk on security, because it's the advantage of edge-cloud integration most easily overlooked.
Architecturally, OpenClaw defaults to running on localhost; background tools all sit on 127.0.0.1. Only one Gateway goes out through an IM long-connection; the outside world cannot touch your machine.
Legally, the edge is a "sale," the cloud is a "lease." Sell cigarettes—cash on delivery and you're done. Run an opium den—you bear accomplice liability. Once an agent can make its own decisions, the side with clear ownership is the safest.
Related Reading
- Is OpenClaw Insecure? You Might Be Even Less Secure
- The Third Argument for Edge AI: Law
- In the Agent Era, Who Is Your Infra Built For?
- Why I Help People Install OpenClaw for Free
On-Site Usage Instructions
- Navigation: Arrow keys / Spacebar / touch swipe / right-side dots to jump
- Editing: Hover over the top-left corner to reveal the ✎ button (or press
E). Click to enter edit mode and modify text directly;Ctrl+Sexports the modified HTML - Fullscreen: Recommended to open the raw HTML directly (the "Open Fullscreen" button in the top-right corner of the page)
The slides for this talk—structure, copy, visuals, illustrations—were all produced in a single session using Claude Code + Gemini. If you want to use an agent to build a deck too, just ask me.