China's "Hard Lobster" Completes Two Rounds of Financing in One Month

When OpenClaw received 250,000 stars in the open - source community and "raising lobsters" became the traffic code on social networks, I learned that a Chinese startup is exploring the implementation form of AI Agents through a different path.

On March 12th, Violoop announced the completion of seed and angel - round financing worth tens of millions of yuan. The financing will be mainly used for product mass production and implementation, global market promotion, and the continuous construction of the Action Model dataset.

Violoop is a technology company focusing on AI PC automation hardware. Simply put, this company attempts to use a small hardware device priced at $250 - $300 to solve the most difficult security and perception problems in the current AI Agent field, so as to better work on behalf of humans 24/7.

Let's start with security, which is a necessity. Shortly after OpenClaw became popular, the macmini, which used to be in long - term stock, started to sell out. When the National Internet Emergency Response Center issued a risk warning, it was sufficient to illustrate the importance of information security at present. It's difficult to solve this problem with software alone. Violoop uses a dual - chip hardware architecture to block security risks. At the same time, in the perception layer, through the combination of software and hardware, Violoop can achieve what OpenClaw can't, enabling AI to "see" the screen and "operate" the computer.

Although it is still in the internal testing stage, the capital market has responded quickly. As I understand it, before the Spring Festival in 2026, Violoop completed two rounds of financing transactions in succession. The first round was completed within two weeks, and the second round signed the SPA (Share Purchase Agreement) in one week. Just recently, I heard from an industry friend that the workload in one week this year exceeded that of a quarter in previous years. You see, this is the financing speed of AI projects at the beginning of 2026.

"When we first started financing, many institutions didn't quite understand what we were doing," the founder Jaylen told me. But when OpenClaw, a project known as the "Linux of the AI era", exceeded 50,000 stars on GitHub, the enthusiasm of the investment market was ignited. "Because there was an overseas benchmark, investors understood what we were doing and thought it seemed very interesting." But as the market heat increased, some investors began to hesitate: Is this thing too big? In other words, if this company really wants to build an operating system for the AI era, it means competing head - on with global technology giants. What are the chances of success for a startup?

The above changes are a typical psychology of current investors when facing AI startups: they are worried about missing out and also worried that big companies will enter the field.

As an entrepreneur, Jaylen doesn't worry too much. With the increasing popularity of OpenClaw, he has already experienced such a mental roller - coaster. In his view, by analogy with the development history of operating systems, after Linux, it seemed that no software operating system could challenge its position, but Windows and Mac still emerged.

"Windows encapsulates a layer on top of Linux and reaches users through killer applications and distribution channels; Mac is an integration of software and hardware, providing a better experience by packaging exclusive products. If OpenClaw is really compared to the Linux of this era, then there will definitely be opportunities similar to Windows and Mac in this era," Jaylen and his team want to explore the opportunities in the infrastructure of the AI era.

Entrepreneurship Triggered by a Fortune Global 500 Project

The birth of OpenClaw was actually full of chance. It was just an open - source project written casually by a retired Austrian programmer to meet his own need of "using chat software to command AI to do work". The same is true for Violoop. It is not chasing hot trends or some grand technological concept, but stems from real business pain points in the process of enterprise operation.

Before founding Violoop, Jaylen had long provided AI technology services to large enterprises, including local model deployment, model fine - tuning, enterprise knowledge base construction, and digital human system development. Most of these customers are global large - scale enterprises with strict internal network security requirements, and all systems need to be deployed in the intranet environment.

In the actual service process, Jaylen encountered a very real problem. Due to the strict access rights of enterprise servers, customers only opened a fixed number of device terminals for Jaylen. And the customer's headquarters is located in Europe, with a time difference. To ensure the stable operation of the system, Jaylen's team members often had to stay up late to monitor the servers to handle unexpected tasks or system maintenance.

"Many times we had to take turns on duty until four or five o'clock in the morning," Jaylen recalled. "The servers run 24/7, but people need to rest."

It was this experience that made Jaylen realize a problem: If AI already has the ability to handle tasks, can it continue to complete some work on behalf of users when they are not in front of the computer?

This thinking became the initial starting point for Violoop's products. Jaylen began to try to use AI Agents to automatically execute tasks. For example, they used the email system to send instructions to the computer, allowing AI to read emails and execute operations. But in practice, they gradually realized that it was difficult to solve some core problems with software alone.

The first is the security problem. In essence, large models are probability systems and may perform unexpected operations in complex environments. For example, automatically deleting files, leaking sensitive information, etc. Once AI Agents are given higher permissions, such risks will be further magnified. If AI can access enterprise servers, it also means that it may accidentally delete key data.

The second is the limited environmental perception ability. Traditional software Agents can only read files or execute instructions, but cannot understand the real operating scenarios of users, such as whether users are browsing the web, processing documents, or conducting data analysis.

The third is the high difficulty of permission management. If the permissions are too low, AI Agents cannot play their roles; but if the permissions are too high, it will bring security risks.

Based on these considerations, Violoop's solution is a dual - chip architecture: the main control chip provides AI operation and acceleration capabilities, and an independent security chip processes execution actions according to a set of rules. Reading and writing files are automatically allowed; deleting files and contacting external contacts require users to confirm through a mobile APP or physical button and release a digital signature before execution.

This architecture also brings another value: By connecting to the user's main computer through the HDMI interface, AI can analyze the screen video stream in real - time frame by frame. For example, when it detects that the user has been organizing invoice photos and extracting numbers into Excel for ten consecutive minutes, the AI will actively ask, "Do you want me to do this?"

The technical basis for this ability is the team's patent accumulation in the field of vision: large - scale video content compression and precise retrieval. Partner King also has rich experience in chip design, which allows them to develop the main control chip at a relatively low cost, thus keeping the selling price at half of that of the macmini.

The price of this hardware is set at $250 - $300, and the subscription service costs $30 per month. The current target customer groups are geeks and small B - end teams, such as design studios, film and television teams, real estate agents, and cross - border e - commerce operators.

A Strategic Location

The reason why "lobsters" are targeted by investors is their entry value.

Investor Bob believes that OpenClaw may become a typical interaction entry. In addition to being a personal external entry, it also has a high degree of automation ability, thus having an irreplaceable experience difference and inclusiveness compared with other interaction entries, and becoming the "entry of entries".

Of course, the primary market's views on the investment opportunities brought by OpenClaw are not unified. Specifically regarding whether to invest or not, a partner of an FA told me that roughly speaking, the entire capital market's views on this matter are divided into two camps. One camp believes that this is a strategic location, while the other camp also agrees with this judgment but thinks more about whether, since it involves the competition for entries, this will be more of an opportunity for big companies?

Some investors explained to me that OpenClaw belongs to the AI orchestration layer, coordinating open - source models and closed - source tools. It is the nervous system connecting AI models and execution tools, a traffic entry and infrastructure. Leading manufacturers will definitely hold this layer in their own hands. Overseas, OpenAI and Anthropic have taken the lead in layout. In this context, there seem to be few opportunities left for startups.

The other view is that due to the incompatibility of ecosystems and the independent operations of big companies, it actually gives a great opportunity to third - parties. Because ecosystems like WeChat and ByteDance obviously will not be compatible with OpenClaw and will probably develop their own claw systems. Under the fragmented ecosystem, the third - parties that can achieve cross - ecosystem compatibility are the most valuable. Especially the teams that can, through technical means, be compatible with a large number of closed - source software without native API interfaces, can break through the last mile that big companies cannot bypass.

"In the AI era, the first - mover advantage is extremely strong. If a creative product idea can be used to break the compatibility problem, an independent path can be taken," the aforementioned partner believes.

Speaking of Violoop, its long - term vision is not limited to being a hardware device. In Jaylen's vision, in the future, everyone will have a dedicated AI, including a dedicated hardware carrier and an edge - side model trained based on personal workflows and data. The current large - model training data does not include details of human - computer interaction such as "how the mouse moves and how the window opens", and their hardware can continuously accumulate user - specific data to train an edge - side model with 4B parameters. Although it is smaller than the cloud - based model with hundreds of billions of parameters, its advantage is that it understands users better and responds faster.

Violoop's understanding of the future operating system is that it is based on tasks rather than files, which requires understanding problems, preparing context, planning solutions, and executing instructions. The edge side is responsible for understanding and data retrieval, and the cloud side is responsible for complex reasoning. Jaylen believes that there is an opportunity to define this new form of cloud - edge combination, although it is still too early to talk about "defining the era". They also hope to equip each AI with an independent email, a virtual phone number, and a digital wallet to serve the AI economic system that will run parallel to the human economic system in the future.

The competitive landscape is changing rapidly. In my opinion, Violoop's advantages lie in its early product implementation, dual - chip security architecture, control ability for software without APIs, and relatively affordable pricing. The current challenge is how to build sufficient technical barriers and user awareness before the giants enter the market.

Although such a product currently seems like a toy that only geeks would buy, we can never underestimate the power of technological change. Because the technological products that have become necessities in people's lives and work today also started as toys in the circle of technology enthusiasts in the early days.

In a recent interview, when the other party asked the founder of OpenClaw why OpenClaw became the winner among so many Agent startups in 2025, the founder Peter said that it was because they were all too serious. "It's hard to compete with someone who is just having fun." The story of Agents has just begun. In a rapidly changing ecosystem, any logical prediction may be wrong, but it doesn't prevent creative people from diving into the wave.

This article is from the WeChat official account "China Venture Capital", author: Liu Yanqiu, published by 36Kr with authorization.