“China's tech titans are embracing an unlikely outsider: OpenClaw, a project created by an Austrian developer that is making waves across the country.
Last week, a line of people queued outside Tencent Holdings' headquarters in Shenzhen, wanting help installing the artificial-intelligence assistant on their computers.
OpenClaw's emergence marks a pivot from previous consumer-facing AI chatbots like OpenAI's ChatGPT and Chinese model DeepSeek, which answer questions, to technology that can also perform tasks.
The open-source AI assistant created by Peter Steinberger can make and carry out decisions on the user's behalf, and has become a hit in China's tech community.
Shares of Tencent rose 7.3% on Tuesday after it launched a suite of OpenClaw-compatible AI products, while shares of startup MiniMax added more than 20% as investors expect it to become a key beneficiary to OpenClaw adoption.
The term "raising a lobster" has been trending on Chinese social media, a nod to OpenClaw's lobster logo as users rush to adopt the AI agent, which can do things such as managing calendars, sending emails and research topics on its own.
Chinese tech companies are capitalizing on OpenClaw's popularity to encourage more people to use their own AI models, while also rapidly building out their in-house offerings.
Tencent's Workbuddy can access popular Chinese office and communication tools and perform tasks at the user's request. TikTok parent ByteDance has ArkClaw, a cloud-based OpenClaw tool that requires no installation, while Alibaba developed an AI agent tool named CoPaw, which supports work messaging apps like DingTalk and Feishu.
OpenClaw also supports a number of foundation models created by smaller Chinese companies like Zhipu AI, which on Tuesday launched AutoClaw, software that makes installing OpenClaw as easy as downloading an app.
Even though the program's popularity could well be a fad, it is lifting investor sentiment toward the sector, Morningstar analyst Ivan Su said.
Sheng Fu, chief executive of Beijing-based Cheetah Mobile, thinks the overall trend will endure, as users increasingly view autonomous agents like "digital employees."
Some Chinese local governments have jumped on the bandwagon, too, promoting the use of the AI assistant as Beijing pushes to develop cutting-edge tech.
Shenzhen's Longgang district announced a draft policy over the weekend that encourages professional platforms to offer free OpenClaw deployment services and provides subsidies for application development.
The high-tech district of Wuxi in Jiangsu province announced subsidies of between 1 million yuan and 5 million yuan, equivalent to $144,774 to $723,871, to encourage industrial use of OpenClaw.
China has been focusing on wiring AI into all industries and fields to create an "intelligent economy," Macquarie analysts said in a recent note.
Still, the capabilities of agentic AI tools have raised some safety concerns. Chinese officials have warned about potential security issues with OpenClaw, given its autonomous decision-making and ability to access system and external resources.” [1]
Can OpenClaw do fine tuning of agent abilities using customer’s video locally?
OpenClaw can perform fine-tuning of agent abilities using customer data, including video, while operating locally. As an open-source, self-hosted orchestration layer, it is designed to run on local hardware (or a private VPS), allowing the agent to learn from, and act on, local data sources and files.
How Local Fine-Tuning Works in OpenClaw:
Real-Time Adaptation: OpenClaw allows for fine-tuning the agent's "personality, memory, and soul" by adjusting model parameters in real-time based on user interaction and feedback.
Contextual Learning: The system uses persistent memory (stored as markdown files locally) to learn user preferences, allowing it to adapt to specific, niche tasks without retraining.
Video/Multi-modal Input: Because OpenClaw can connect to local vision models (such as via Ollama or Clarifai Local Runners), it can process video or image inputs to inform its actions.
Self-Improving Skills: OpenClaw agents can be instructed to "self-improve" by analyzing previous interactions, effectively creating new, tailored skills from experience.
Key Considerations:
Security: Because OpenClaw operates with high-level access to your local file system, it is highly recommended to use sandboxed environments (like Docker) or isolated VMs to prevent unauthorized access.
Technical Skill: While it allows for powerful local customization, it requires comfort with terminal commands, setting up node.js, and managing API keys.
How to set up isolated VMs on my computer and how much does it cost to use them?
Isolated VMs (Virtual Machines) are set up for free locally using hypervisors like VirtualBox or Hyper-V by installing them, downloading an OS image (e.g., Linux ISO), and configuring virtual hardware (RAM/Disk). For isolation, configure network settings to "Host-only" or disconnect network adapters. Cloud options (Azure/AWS) cost roughly $5–$10/month per VM, while local setup is free.
How to Set Up Isolated VMs Locally (Free)
Using software on your existing computer is the best way to ensure free, secure isolation.
Software Options: VirtualBox (Windows/Mac/Linux), VMware Workstation Player (Windows/Linux), or Hyper-V (Windows Pro).
Steps:
Download & Install: Download VirtualBox or VMware, then download a guest OS ISO file (e.g., Ubuntu Linux). Always download Windows ISO files directly from Microsoft to avoid compromised third-party versions. You need at least 5 GB of free space and a 64-bit or 32-bit CPU to create the ISO, depending on your target system.
Create VM: Open the software, click "New," name your VM, and select the downloaded ISO.
Allocate Resources: Assign at least 2GB RAM and 20GB disk space for basic usage.
Network Isolation: To isolate, go to VM settings -> Network, and change attached to "Host-only" or disable the network adapter completely.
Install OS: Start the VM and follow the on-screen instructions to install the operating system.
1. China's OpenClaw Craze Fuels Stocks, AI Pivot. Qin, Sherry; Qu, Tracy. Wall Street Journal, Eastern edition; New York, N.Y.. 11 Mar 2026: B4.
Komentarų nėra:
Rašyti komentarą