After the Lobster Hype Cooled, HONOR Built It a Universe

OpenClaw doesn't need more applause. It needs a hardware ecosystem where it can actually grow.

👦🏻 Author: GaKi

🥷 Editor: Koji

🧑‍🎨 Layout: NCon

Every technology wave follows a similar rhythm: first it blows up, floods your feed, gets mythologized; then the hype recedes, leaving behind the handful of people still willing to keep refining it.

OpenClaw has reached that moment.

After the heat fades, the stuff that actually matters starts happening. What it needs now isn't more applause — it's an environment where it can grow long-term: lower barriers to entry, more stable experiences, controllable privacy, and a more reasonable cost structure.

This happens to be exactly where hardware makers are most likely to step in.

And it's at precisely this moment that HONOR, at its April 13 PC product tech briefing, announced a product combo that looks pretty interesting:

The HONOR Lobster Universe, YOYO Claw — a "raise-your-lobster" ecosystem that embeds OpenClaw into PCs, tablets, and phones.

Then on April 16, the first "lobster-raising laptop" equipped with it — the new MagicBook — officially went on sale.

If the "AI does it itself" paradigm that OpenClaw pioneered can smoothly enter the HONOR ecosystem, what it brings will likely be a shift at the level of user habits.

🚥

The Crossing team spent a few days testing it out. Here's what we found.

What exactly is the YOYO Claw Lobster Universe?

The core structure of YOYO Claw Lobster Universe can basically be explained in one sentence: on each of HONOR's three endpoints — PC, tablet, and phone — there's a lobster built in, and each endpoint can raise lobsters.

To see the full logic of this product, the PC side is the most direct entry point. We specifically got in touch and purchased a HONOR MagicBook Pro for testing.

After logging into YOYO Claw on the PC side, the first impression is the simplification of the interface layer. The command-line-style conversation that OpenClaw uses in terminal, here gets compressed into something closer to a chat tool interface. At this layer, you can directly configure personality, character, and other parameters for your lobster.

The system already has a batch of pre-built lobsters for different scenarios, covering PC usage, self-media operations, document processing, and other directions.

According to official specs, the current HONOR YOYO Claw comes with 5 pre-installed main lobsters and 23 sub-lobsters out of the box, covering 5 scenarios: education, office work, academics, content creation, and intelligent assistance.

From the user perspective, no API configuration, no model routing setup, no Skill writing — basically a ready-to-use state.

Beyond the built-in lobsters, you can find more in its Bot marketplace, with a simplified configuration flow — just click to add.

In the main lobster, commonly used capabilities are already pre-configured. General modules like image understanding and image generation, plus their corresponding Skills, are basically set up — no need for users to install them separately.

Because the whole system is built directly into HONOR PCs, the lobster's permissions for local machine operations run pretty deep. For example, downloading and installing common software like Steam or QQ — it can open websites, call tools, execute installation via command line, all without user intervention.

Developer-oriented tools work too. Having it install Claude Code locally and configure the dependency environment — it runs through the whole process itself.

It can also go directly to GitHub to search for specific projects, like CCC Switch — it grabs the project page first, then executes installation.

Local storage management can be handed to it too. For example, having it scan the C drive, list files taking up significant space, then letting the user decide what to clean up.

For research-oriented tasks, the lobster can complete dozens of rounds of web search and tool calls in sequence.

In testing, we gave it a relatively deep research task: mapping out notable AI Agent startups in interview scenarios, domestic and international. The research ultimately generated a report of over 10,000 words.

Because it has Agent properties itself, this research content gets automatically organized into a formatted, relatively complete Word document, saved directly to local storage.

Coding tasks work too. For example, having it generate a multi-card style, fully laid-out HTML showcase page — the completion level is decent:

Stringing these scenarios together, HONOR PC-side YOYO Claw has a fairly obvious product orientation: push the usage barrier as low as possible.

Because the system comes factory-installed, with all Skills and capability modules pre-configured, this setup lets it run smoothly in relatively lightweight scenarios like home and education. One concrete example is grading in education scenarios: for relatively simple multiple-choice image questions, it delivers acceptable recognition accuracy.

In testing, we told it there were several screenshots on the desktop that needed grading. It first found the images locally, categorized them, then performed 14 rounds of tool calls in sequence, executing image recognition tasks one by one.

After one round of recognition, it self-checks the content. If it finds missing information, it runs another round of recognition. All content ultimately gets written into a Markdown document first, which can later be reorganized into a doc document — overall processing speed is decent.

The grading report in the DOC document gets basic formatting treatment. One thing to note: because the underlying layer is AI doing image recognition, structured questions like multiple choice work better, while open-ended problems are more challenging to interpret.

In YOYO Claw, HONOR PC natively supports remote phone control. The system has already built in several connection paths — the pairing process is basically two clicks to connect.

For example, after scanning a QR code in a WeChat scenario, you can quickly create a WeChat ClawBot on the phone side.

Another design worth mentioning: HONOR PC-side YOYO Claw supports connecting to local large models. This is a point HONOR PC repeatedly emphasized in its introduction.

At the public information level, one number worth recording: compared to directly using the community version of OpenClaw, YOYO Claw's comprehensive Token consumption on PC drops by approximately 50% — this data comes from HONOR's official disclosure. The mechanism behind it is an "edge-cloud collaboration, edge-first" routing approach.

The logic roughly works like this: simple tasks are handled on-device; complex tasks go to the cloud.

Tablet

In the YOYO Claw universe, PC was actually the last endpoint to be added. The tablet came first, then phone — PC was the last piece of this system to land.

On the tablet side, HONOR provides runtime support through a Linux desktop environment at the底层, with a one-click deployment script pre-installed. On HONOR tablets, users can basically complete OpenClaw deployment through a few clicks.

Opening YOYO Claw on a HONOR tablet, the deployment process is essentially simplified to a few clicks. Unlike the PC community version where you manually configure environments and install dependencies, the system has already packaged the OpenClaw runtime environment for users.

The specific usage path HONOR designed for this is: through the Pad-side developer mode, enter the Linux laboratory's built-in desktop to perform installation.

The direct significance of this: the debate that raged for a long time in developer circles about "can tablets run Agent" — here gets a relatively complete answer.

The lobster running on tablet has the same capability set as the PC side at its source, but constrained by the tablet's own computing power and usage scenarios, it leans more toward lightweight tasks: local file organization, note categorization, simple image and text generation. Its relationship with the PC side is more like "the same lobster in different forms on different devices."

Phone-side YOYOClawPhone

The phone side is something relatively different in HONOR's layout this time. It's fundamentally different from the lobsters on PC and tablet.

The lobsters on PC and tablet rely more on remote connections. YOYOClawPhone Pioneer Edition can run directly on Android phones.

This means it's more like a true on-device lobster. The phone is no longer just an entry point for sending commands — it can itself call apps, files, and device capabilities to complete specific operations.

Its barrier to entry is also lower: one-click install on Android phones, interface more friendly to beginners; Skills, memory, and data stay local, not uploaded to cloud. Beyond routine tasks like information organization and content processing, it can directly operate the phone and apps, handling small life tasks in a "zero-touch" way.

It's more like building bridges between different devices. Whether Android phones, PCs, tablets, or other smart hardware — as long as it can connect to this system, it can be mobilized by different "lobsters."

Tasks no longer rely on a single Agent either — they can be split among multiple roles working together. This way, hardware is no longer just equipment waiting for commands; they're assigned roles and start participating in a collaborative AI network.

It divides AI Agents into several more easily understood roles: ones that help you get things done, ones that chat with you, and ones that lean more toward emotional companionship.

This design is a bit like giving "lobsters" character profiles. They're not just responsible for executing tasks; they also try to create some familiarity and companionship for users during use. Put more bluntly, it's about raising a tool lobster into a lobster that understands you better.

Simply put, what YOYOClawPhone wants to do isn't add another AI assistant entry point to phones — it's to put the "lobster" directly into the phone.

It can run locally on Android phones, call apps, files, and device capabilities in the phone, and also interact with other hardware. This way, the phone isn't just a remote control for sending commands — it starts becoming an AI device that can execute tasks itself.

The significance of this is that OpenClaw is no longer just revolving around PCs. It's starting to enter mobile, and making "a lobster for every person" something closer to ordinary users' daily lives.

YOYO Lobster Control

Above these three endpoints, YOYO Lobster Control is a relatively independent capability in this system. Its core logic is: phone as entry point, remotely connecting to and commanding the lobster on PC to execute tasks.

The connection process was already mentioned in the PC testing section — the system has built in several association paths, basically two clicks to connect. After scanning a QR code in a WeChat scenario, you can quickly create a WeChat ClawBot on the phone side, serving as a communication channel between phone and PC.

After connection is complete, a two-way channel forms between YOYO on the phone side and the lobster on the PC side.

Here are a few typical cross-device scenarios:

【1】Phone sends commands, PC executes.

【2】The lobster on PC calls phone-side capabilities (SMS, contacts).

【3】Scheduled tasks and proactive push notifications.

Stringing these scenarios together, the positioning of YOYO Lobster Control is fairly clear:

In this structure, the phone plays a dual role of entry point and remote control. You don't need to walk up to each device to operate it — just issue commands on the screen you hold most often, and lobsters distributed across PC, tablet, or even third-party clouds will each execute and return results.

According to official introduction, YOYO Lobster Control's compatibility isn't limited to HONOR's own devices — it also supports connecting to OpenClaw deployed on third-party clouds and third-party devices. That is, as long as the protocols match, one phone can uniformly schedule multiple lobsters from different sources.

Put simply, what YOYO Lobster Control wants to do is let AI stop at the "you ask one question, it answers one question" stage. You say one sentence on your phone, and it can mobilize PCs, tablets, even lobsters elsewhere, completing a chain of tasks in sequence.

It's targeting the problems that most easily turn people away from AI: tasks that don't flow coherently, configuration processes that are too much hassle, high costs from long-term use, and the lingering unease after granting permissions.

At this point, there's a question that can't be avoided: the more capable the lobster is, the more things it can touch. Being able to organize desktop files also means it has permission to read your private information on the desktop.

This is why the better OpenClaw-type things get made, the more concerns people have. YOYO Claw has done quite a bit at this layer.

For example, core data is only processed locally. All personal memory, sensitive data stays on the device, not going to cloud. There's even something called a device-level independent security lobster that monitors other lobsters' AI operations — things like formatting hard drives, reinstalling systems, the lobster won't do.

Other sensitive operations like payments, opening cameras, etc. require secondary confirmation.

After OpenClaw blew up, the real problems are just beginning

One of the exciting things about OpenClaw when it blew up was: it might be the first time ordinary people could know how to make AI "do things itself."

But anyone who's actually deployed and used it for a while would probably agree with one statement: after the novelty wears off, OpenClaw is still at the "good toy" stage — there's still a distance to "assistant."

There are probably at least three things that will gradually wear down your patience.

【1】After installation, you need to raise the lobster.

OpenClaw itself is open source, so in theory anyone can run it. But in actual operation, environment configuration alone will turn away a large batch of people. This barrier can spread in programmer circles, but it's hard to truly spread to ordinary users — which is why many manufacturers are entering the space.

Secondly, after installation, you still need to raise it. At this step, ordinary users don't have much of a "raising" mindset.

【2】Tokens are too expensive.

OpenClaw's working method is to let the model continuously "remember, think, act, think again." A slightly complex task, running for half an hour, might cost ten or dozens of yuan.

We have friends around us who said, in the first week after setting it up, they checked the dashboard every day. And this characteristic of OpenClaw even led Claude and other manufacturers to directly ban using membership subscription quotas to run OpenClaw.

【3】Privacy concerns.

Handing over desktop control of your computer to a product running in the cloud — there's bound to be some concern.

These three things combined lead to a very common situation: OpenClaw gets passed around like crazy on social media, but if you open the computers of ten people around you who've used it, probably eight haven't clicked on it in a month.

Looking back, this is actually the necessary path for any new technology going mainstream.

What OpenClaw lacks right now isn't stronger capabilities — it lacks a "home."

We can actually draw an analogy with Linux. The Linux kernel came out in 1991, but what really brought it into millions of households wasn't the kernel itself. What actually made the difference was things like Ubuntu, which packaged it into something "ready to use out of the box."

Without Ubuntu, Linux might have forever remained a hacker's toy; with Ubuntu, it had a chance to become an operating system ordinary people could use.

What OpenClaw lacks now is basically something like this "Ubuntu."

So the question becomes: who will make OpenClaw's Ubuntu?

Intuitively, the players most suited for this are actually phone and PC manufacturers. Because they simultaneously hold three things: hardware, operating system, and users.

And HONOR, this phone manufacturer, is what appeared in this context and gave the tentative answer we tested above: the YOYO Claw Universe.

Zooming out: phone manufacturers and AI capabilities are undergoing an important collision

So, what route is YOYO Claw taking among phone manufacturers?

Over the past half year, the capability of "AI doing things itself" has started migrating from open-source frameworks like OpenClaw that lean toward developers, toward mass consumer products. One relatively conspicuous milestone was December 2025, when ByteDance partnered with ZTE to launch the Doubao AI phone engineering sample — the first batch of 30,000 units sold out instantly, with secondary market premiums exceeding 3,500 yuan at peak.

The most watched capability of this phone was system-level GUI Agent — that is, AI directly reading the screen, simulating clicks, completing coherent tasks across applications.

It pushed one thing to the industry forefront: when a super Agent can bypass app sandboxes, directly see the screen and directly tap the screen, the ceiling of capability and the floor of privacy collide head-on.

After the debate, two routes gradually emerged in the industry.

【1】System-level GUI Agent.

【2】On-device + A2A (Agent to Agent)

Beyond these two routes, there's also a slowly forming consensus in the industry: for GUI Agent to truly scale, it can't bypass "on-device-ization."

On-device AI Agent capability isn't a performance concept — it's the prerequisite for GUI Agent to go mainstream.

Google is putting more personal data computation back on-device through Private Compute Core; OPPO is pushing "Memory Symbiosis Engine" in ColorOS, data not going to cloud; Apple has similar moves at the system level.

Putting YOYO Claw back in this picture, HONOR's position this time becomes fairly clear.

It didn't choose the aggressive system-level GUI Agent route, nor did it go pure protocol-level A2A. Its entry point landed on the on-device AI branch: reorganizing the OpenClaw open-source framework on the terminal side.

Looking longer term, what YOYO Claw echoes is HONOR's two higher-level formulations: the Alpha Strategy "three-step plan" (smartphone, smart ecosystem, smart world), and AHI (Augmented Human Intelligence, or "wisdom that puts humans first").

These two concepts sound somewhat abstract, somewhat grand, but when landed on this specific product combination of YOYO Claw, the goal can actually be reduced to one relatively朴素 wish:

Make the AI Agent thing safe, convenient, economical, and usable by ordinary people in the long term.

Placed in the context of the on-device Agent route debate, this choice counts as a relatively complete attempt by a phone manufacturer at on-device AI.

🚥

YOYO Claw still has its own problems to solve, maturity still needs time to polish, and whether the three-endpoint interconnection can achieve smoother, more frictionless interaction.

This on-device Agent route still needs to be validated by real scenarios and real users — there's no ready answer yet.

But looking back at AI these past few years, from large models, to Agent, to on-device execution — perhaps the true "a lobster for every person" era doesn't start from a lobster-raising laptop.

It starts from the day we finally dare to hand our computers, tablets, and phones over to a safe, understanding, capable little lobster.

And this, might be how AI truly starts serving everyone in 2026.

Crossing is looking for independent contributors to write AI product and model reviews.

If you've written similar articles: "Hands-on: PixVerse C1", "Hands-on: LibTV", please contact zeo0811@gmail.com. Email should include: ① personal introduction, ② AI review articles you've written.

We offer competitive compensation. Looking forward to observing and recording the AI era with you 🎪