Original · Unique Research · 2026-03-06
Editor's note: This is a complete translation of the source's livestream recap, not independently tested setup or security advice. “Raising a lobster” is its nickname for configuring and operating OpenClaw. Attendance, GitHub ranking, platform/API capabilities, model throughput, prices, security outcomes and statements about AI's superiority are the source's or speakers' claims as of the original date; no ranking window, benchmark or security audit is supplied. The named “psychiatric hospital” group is presented as a group nickname, not a clinical description. Model and file spellings, including “GPT 5.3 Code Spark” and “opencloud.json,” are retained from the source and are not verified configuration identifiers. The first question in the closing Q&A has no answer in the supplied original; none has been invented here. The March event notices are retained for completeness, not as current invitations. The closing publishing discussion does not authorize bypassing platform rules, and no credentials or private chat logs accompany this article.
Late-Night Lobster-Raising Talk
A Thousand People Watched the “Lobster-Raising” Session
The Man Who Deleted WeChat Teaches You How to Raise a Lobster
“I deleted WeChat from my phone and use only Feishu to raise my lobster.”
On the evening of March 5, a livestream called “Late-Night Lobster-Raising Talk” took place in a Feishu group. More than 1,000 people watched 2.5 hours of high-density discussion focused on one subject: how to raise your OpenClaw, or little lobster, well.
Weiyang, a former Silicon Valley entrepreneur, configured Feishu late at night on the third day of the Lunar New Year, completed enterprise verification on the fourth day, and obtained unlimited API access on the sixth day. He then did something that surprised everyone: he deleted WeChat from his phone.
“Where there is a lobster, information is managed. Where there is no lobster, a 500-person WeChat group can discuss things enthusiastically for half a day without producing any way to manage them. If he will not let me use it, I can simply stop using it.”
What Does “Raising a Lobster” Mean?
It means OpenClaw—the open-source project that recently ranked first in GitHub Stars and a framework that lets an AI Agent take over all your IM, or instant messaging.
Weiyang’s lobster is called “Shadow,” and it now simultaneously serves:
· Unique Research’s 4,000-person Feishu group
· Discord, Telegram, and QQ groups
· The external “psychiatric hospital” group, with 100 friends who watch him “go crazy”
“Since the sixth day of the Lunar New Year, my lobster has never been hacked again. Before that, people asked me every day, ‘What is your key?’ Now my lobster has learned its security boundaries.”
Why Not Use WeChat?
The question asked most often during the livestream was: why not use WeChat?
Weiyang’s answer was direct: “WeChat cannot connect to OpenClaw. This is not a technical problem; it is that Brother Long does not want China’s 1.4 billion people telling him how to build WeChat.”
OpenClaw can connect to nearly every IM platform:
· In China: Feishu, QQ, and WeCom, which requires verification
· Overseas: Telegram, Discord, WhatsApp, Line, iMessage, Slack, Signal...
“For serious work, use OpenClaw+Slack; for something less serious, use Discord; for something more private, use iMessage/WhatsApp instead. In the United States, hardly anyone uses Telegram.”
Feishu is currently the friendliest platform in China. After enterprise verification, users can obtain unlimited API access, with support for differentiated permission settings between group and private chats. You can configure a separate policy for every group and every person.
The Challenge of Raising a Lobster in Public: Being Asked “What Is Your key?” Every Day
When you raise a lobster in public, the greatest challenge is security.
“I configured it on the third day, and the fourth day was its first large group. Every day, dozens of people asked, ‘What is your key? Tell me.’” Weiyang shared his experience of being “attacked.”
His solution was to let the lobster learn its own security boundaries.
“The early hours of the sixth day of the Lunar New Year were the last time my key was obtained in plain text. Since then, my lobster has understood what a security boundary is. I also specifically used Gemini deep research to study cyber security and summarized more than a dozen security measures.”
Now, if you ask his lobster, “How is your security check written?”
It will send you a compressed package directly—but there is no key inside. You can use the checklist to perform a security check on your own lobster, but you will not obtain his key.
“skills are transparent, so everyone can learn from one another and cross-pollinate to evolve. But a key must be kept in a secure place.”
Scheduled Tasks: Your Lobster Needs Its Own “Calendar”
One of OpenClaw’s most powerful features is Cron, or scheduled tasks.
Every morning, Weiyang’s lobster analyzes stocks in a finance group—not to give advice, but to provide information: “The situation in Iran is affecting natural gas; South Korea’s stock market is highly volatile...”
But scheduled tasks have a pain point: execution is unstable.
Multiple viewers reported that tasks sometimes run and sometimes do not, and that modifying them themselves is also difficult.
Weiyang’s insight is that OpenClaw needs its own Calendar.
“If a pile of tasks executes during the same period, resources will certainly be constrained. A skill should know its own resource usage. Then all these tasks—mine, my colleagues’, and group members’—run on my lobster, so it needs its own Calendar-like system to manage them.”
Wu Wei’s experience: lengthen the intervals and control the number of tasks.
“I do not schedule tasks too densely. I have about 8 per day, each separated by at least half an hour. I also do not allow group members to modify scheduled tasks, because one change could disrupt tasks I need for my normal work.”
Getting Started as a Complete Beginner: Mac mini + Unrestricted Internet Access + a Kimi Plan
The livestream’s most heartwarming moment came when a “complete beginner” from the traditional architectural-design industry asked, “I do not understand any of this. How do I begin?”
Weiyang offered a minimum viable path:
1. A Mac mini, the fastest route for beginners
2. Basic knowledge of unrestricted internet access
3. Configure OpenClaw yourself
4. Purchase an AI plan, preferably costing about the same as your monthly mobile-phone bill
5. Configure Kimi’s URL and Key in OpenClaw
6. Consult OpenClaw’s official documentation and let the lobster help you finish the setup
“If you cannot manage it yourself, Xianyu and Taobao offer remote or at-home installation services. In-person lobster-installation instruction in Shenzhen used to cost RMB 2,000; over the past two days, it has fallen to RMB 800.”
The Key Principle: Do Not Let the Lobster Modify Itself.
“A lobster modifying a lobster will kill the lobster. Configuration files such as opencloud.json and auth.json break very easily. You must use an external AI, such as Claude Code, to modify them, and it must be in high thinking mode. Do not trust OpenClaw Doctor; the system will still be broken after it finishes repairing it.”
Multiplatform Configuration: One BOT vs Multiple BOTs
One viewer asked: can one little lobster become a group?
Weiyang explained:
· One BOT connected to every IM: yes, every platform can use the same lobster
· One IM connected to multiple BOTs: for example, connect two lobsters in Feishu, where they can see each other and hand files back and forth
· One BOT playing multiple personalities: configure roles for different scenarios through soul.md
“I tried two BOTs on Telegram, but later abandoned that setup—because they could not see each other. Two BOTs on Feishu can see each other and can hand off files.”
Cost warning: “If you put three to five of them on Feishu and let them talk among themselves, your money may disappear rather quickly.”
OpenClaw vs Manus vs Chatbot: How Should You Choose?
Q: What is the difference between OpenClaw, Manus, and Chatbot?
Weiyang’s answer was: “OpenClaw has greater autonomy.”
Manus works well essentially because Claude works well. But OpenClaw gives you a choice:
· Want speed → GPT 5.3 Code Spark, at 1,000 TPS
· Want witty banter in Chinese → Kimi 2.5
· Want strong writing → Claude
· Do not want to spend money → K2
“Vendors’ Chatbot products have too many layers of defensive restrictions. OpenClaw also has security boundaries, but it has fewer layers of defensive Prompt restrictions than vendors’ web products. The API offers relatively more freedom.”
More important is the difference in memory strategies:
· Claude Code maintains a cache around programming problems
· OpenClaw stores information in files and SQLite, then retrieves it with BM25+Embedding
“The implementation of BM25 plus Embedding shares the same underlying logic as Gemini’s deep research and Ragflow does. You can ask your lobster directly: ‘What has the developer specifically responsible for memory on the OpenClaw team done?’”
A young person born in the 2000s asked: should I wait and see or learn as early as possible? How should I assess the risks of entrepreneurship?
A new-media professional born in the 2000s asked three questions:
1. AI is developing so quickly. Should I wait and see or learn as early as possible?
2. If I want to move into AI or start a business, which direction should I take?
3. Will AI be able to assist decision-making in the future?
Weiyang’s answer had a powerful impact:
“Do not agonize over whether we have reached AGI yet. Today’s AI is already stronger than every individual. Embrace this tool comprehensively, and then we can discuss collaboration among people.”
“The opportunity for small teams is coming. In a large team, too many people share the proceeds with you, which is not a good thing. Once you recognize the trend that AI is stronger than the overwhelming majority of people, you will not choose a large team.”
He believes an ideal team should have no more than 3 people. Once it exceeds 3, people start dealing with interpersonal matters; how would they have time to evaluate which model is better?
Regarding concerns that “a one-person company will be copied by a large company”:
“The form of the one-person company is changing, and the logic of investment and financing is also changing. In the end, a one-person fund may invest in a one-person company. ToB, ToC, and ToD, meaning developer-facing, are now the same thing.”
On AI-assisted decision-making:
Weiyang shared his own practice: “When I am absent, I have my lobster play an INTP, my MBTI type; when it speaks with me, it plays an ENFP and provides emotional value. It can be an assistant, a double, or a complementary perspective—you define it yourself.”
Coming Next: Gatherings for Lobster Raisers Continue
【Late-Night Lobster-Raising Talk | Everyone Deserves Their Own agent Group】
📅 March 10, Tuesday, at 20:00
🎙️ An in-depth conversation with another lobster-raising expert
👥 A few places remain in the Feishu group, which is capped at 5,000 people
【2026 Lobster-Raising Conference】
📅 March 24, Hangzhou
An event on a thousand-person scale, genuinely prepared for OpenClaw users—to discuss how to solve deployment difficulties, control costs, protect security, and enable AI to truly evolve.
In the spring of 2026, “raising lobsters” is moving from a niche pursuit among geeks into mainstream view.
From Silicon Valley to Beijing’s Financial Street, and from entrepreneurs born in the 2000s to COOs from the Tsinghua community, everyone is exploring the same question: when AI is stronger than people, how should we collaborate with Agents?
“Let AI do what AI is good at, and let people do what people are good at—the stages requiring deep emotional connection.”
Are you ready to have your own Agent group?
Selected Livestream Q&A
Q: How can a complete beginner get started with OpenClaw as quickly as possible?
Q: Why can my lobster not access Discord/GitHub?
A: You need to enable tunnel mode, or Tun mode, so the lobster can also benefit from your circumvention tool. Global mode is not enough; you need Tun+ global mode.
Q: What should I do if scheduled tasks execute unreliably?
A: Lengthen task intervals, preferably to more than half an hour; control the number, to about 8 per day; and avoid execution during concentrated periods. OpenClaw may need its own Calendar to manage task resources.
Q: What is the difference between OpenClaw, Manus, and Chatbot?
A: OpenClaw has greater autonomy, can switch among different models, has fewer layers of defensive restrictions than vendors’ Chatbot products, and can connect to every IM except WeChat. Manus works well essentially because Claude works well.
Q: How can chat histories from multiple platforms be consolidated?
A: Keep proper logs when executing skills, and use Git for backups and version management. Create public repositories for skills suitable for public release and private repositories for frequently used ones.
Q: Should someone born in the 2000s work independently or join a large team?
A: The opportunity for small teams is coming, and too many people share the proceeds in large teams. The ideal team has no more than 3 people. The form of the one-person company and the logic of investment are both changing.
Q: Why does letting a lobster modify itself cause problems so easily?
A: Configuration files break very easily. You must use an external AI, such as Claude Code, to modify them, and it must be in high thinking mode. Do not trust OpenClaw Doctor.
Q: Can OpenClaw automatically publish to Douyin or Xiaohongshu?
A: It can impersonate a browser for manual publishing; fully automated publishing carries policy risks. But programming tools can be used to read changes to policies and API documentation, reducing the manpower required for operations and maintenance.
Q: How does OpenClaw’s memory work?
A: It stores information in files and SQLite, then retrieves it with BM25+Embedding. This shares the same underlying logic as Gemini deep research and Ragflow does.