QClaw

• 1. overview (820 chars)

  QClaw is a localized AI Agent assistant launched by Tencent's computer manager team in early 2026. It is developed based on the open source OpenClaw ecosystem and focuses on the unique experience of "remote computer control via WeChat". The product adopts an architecture design that separates the control plane and execution plane. Users can remotely schedule local computers to perform complex tasks such as file sorting, code generation, and browser operations through WeChat. Currently in the public beta stage, the client is completely free and provides a daily computing power subsidy of approximately 40 million Tokens. QClaw's core competitiveness lies in transforming AI Agent from a "geek tool" into a "productivity tool readily available to ordinary users." Relying on WeChat's access advantage of 1.4 billion users, it has established differentiated barriers in terms of local data security, zero-threshold deployment, and continuous memory learning.

• 2. Product background (1186 chars)

  QClaw was developed by Tencent Computer Manager team. It officially launched internal testing in March 2026. It launched the V2 major version on April 9 and started full public testing. The product nickname is "Crayfish AI" and is built based on the OpenClaw open source framework. It is an important layout of Tencent in the AI ​​Agent track.   From a strategic perspective, QClaw forms a complementary product matrix with ClawBot within the WeChat ecosystem and WorkBuddy for corporate office scenarios. ClawBot is positioned as a conversational AI assistant within WeChat, WorkBuddy is aimed at enterprise-level office automation, and QClaw focuses on local Agent control on personal computers. This layout reflects Tencent’s strategic intention of “grabbing the entrance” – leveraging WeChat’s massive user base to allow AI capabilities to reach ordinary users at the lowest threshold.   Tencent’s investment in the field of AI Agent is not an isolated act. From 2025 to 2026, major domestic manufacturers have launched their own "lobster" products, including Alibaba Cloud's Claw, ByteDance's Button, Baidu's Miaoda, etc., forming the so-called "Battle of One Hundred Shrimps." In this competitive landscape, QClaw has chosen a differentiated path: it does not pursue "technical brilliance" that pursues general intelligence, but instead focuses on the pragmatic route of "solving real needs."

• 3. Product functions and usage experience (239 chars)

  QClaw’s core capabilities can be summarized as “let WeChat become your computer remote control”. After the user completes a simple scan code binding, he or she can send natural language instructions to the computer through WeChat on the mobile phone to realize remote task execution.

• 4. Core Function Highlights (2318 chars)

  WeChat direct connection remote control is the most recognizable feature of QClaw. Users do not need to configure environment variables, Docker or API Key, and initialization can be completed in three steps after downloading and installing. After binding WeChat, no matter where you are, you only need to send commands in WeChat, and the local computer will automatically execute and feedback the results. This kind of "intent-level scheduling" is different from traditional "pixel-level remote control". Users do not need to accurately describe the operation steps. They only need to describe the task goals, and QClaw will automatically disassemble and execute them.   Local deployment and data security are another big selling point. QClaw runs entirely locally on the user's computer, and task execution, file processing, and data storage do not leave the device. This "sovereign agent" architecture combined with native security sandbox protection is highly attractive to privacy-conscious users and enterprises. QClaw's local-first strategy eliminates concerns about data leakage compared to solutions that upload data to the cloud for processing.   Multi-model support reflects the openness of the product. QClaw has built-in domestic high-quality large models such as Kimi, MiniMax, GLM, and DeepSeek, and also supports user-defined access to third-party model APIs. This "decoupling of model layer and execution layer" design allows users to flexibly switch models according to task requirements, and also provides enterprise users with more compliance options.   Continuous memory and personalized learning are vividly called "shrimp farming". QClaw can remember the user's usage habits, preferences and historical context, and will gradually understand the way the user works over time. For example, it can learn the user's email writing tone, file archiving preferences, etc., to achieve an experience that "the more you use it, the better it gets to know you."   Major upgrades to the V2 version further strengthen product capabilities. The multi-Agent collaboration function supports the creation of up to 3 Agents working in parallel, each with customized expertise, skills, and permissions to achieve the disassembly and division of labor of complex tasks. The application "connector" function is connected to mainstream applications such as Tencent Documents, Tencent Meetings, Kingsoft Documents, Notion, and Mailbox, and the steps for a single task are expected to be reduced by more than 60%. The "Lobster Butler" security protection function can open a safe environment with one click and block risks such as malicious prompts, skill poisoning, accidental file deletion and sensitive information leakage.

• 5. User experience evaluation (1114 chars)

  According to 36 Krypton’s actual test evaluation, QClaw’s installation experience received a perfect score of 10 points, and the deployment time did not exceed 3 minutes, making it truly “ready to use out of the box”. However, the user experience score is 6 points, and the reviewer believes that "it is currently more like a running demo than a usable product."   Functions that passed the test in the actual test include opening the browser, entering the URL, jumping, and installing and opening the software. Functions that failed the test include searching for and sending articles, helping users send WeChat messages, etc. There was a delay of about 2 minutes in task reminders. The main problems focus on the lack of domestic localization adaptation, the need for users to manually configure the environment, and the deviation in the understanding of Chinese instructions.   However, QClaw’s iteration speed is impressive. Since the full public beta, the product has iterated on more than 30 functions in a short period of time. The release of the V2 version marks the product's progress from "available" to "easy to use." Official documents show that QClaw currently has more than 5,000 community skills, covering multiple scenarios such as office, creation, development, etc., and the skills ecosystem continues to grow.

• 6. Pricing and business model (1629 chars)

  QClaw currently adopts a "free + subsidy" business model. The client software is completely free, with no software purchase fees and no annual subscription fees. Users can directly download the Windows or macOS version installation package from the official website.   In terms of computing power costs, the official provides a free quota of approximately 40 million Tokens every day. Basic conversations consume approximately 18,000 Tokens each time, and complex tasks consume approximately 180,000 Tokens each time. According to this standard, the daily free quota is enough to perform 200 to 300 complex tasks, and ordinary users can basically achieve "freedom of shrimp farming".   If users choose to switch to a third-party large model, they will be responsible for their own API fees. QClaw supports access to large models from other manufacturers such as Tongyi Qianwen and Doubao. Users need to apply for the API Key of the corresponding platform by themselves and deduct fees in real time according to the third-party manufacturer's rate standards. In this case, the QClaw software itself is still free of charge, but the computing power cost is borne by the user.   In the long term, QClaw’s business model may evolve toward a subscription-based model. Referring to the pricing strategy of similar products, Huawei Xiaoyi Claw is priced at 49 yuan per month, and Kimi’s subscription service is priced at 199 yuan per month. Tencent accumulates users through computing power subsidies during the public beta stage, and may launch paid value-added services or subscription plans when the product matures.   It is worth noting that "door-to-door installation" or "paid uninstallation" services with prices of several hundred yuan have appeared on social media. Officials have clearly reminded users to be wary of such third-party charging traps and recommend downloading and installing independently through official channels.

• 7. User reviews (1368 chars)

  User reviews of QClaw are obviously polarizing. On the one hand, professional evaluation agencies have given high evaluations. In the horizontal evaluation of 10 domestic intelligent agents, QClaw tied for first place with another product. On the other hand, real user feedback includes complaints such as "slow response", "self-assertion" and "message packet loss".   Supporters believe that QClaw has successfully lowered the threshold for using AI Agents through the WeChat portal and local deployment, truly achieving inclusive technology. For ordinary users who don’t understand code, being able to deploy and start using an AI automation tool in 3 minutes is a breakthrough in itself. The convenience of WeChat as an entry point has also been widely recognized. Users do not need to learn a new operating interface and can complete complex tasks in familiar chat scenarios.   Opponents are concerned that the product is still in its early stages and its stability and reliability need to be improved. Some users have reported that QClaw will exhibit "hallucination" behavior when performing tasks, that is, misunderstanding the user's intentions or performing wrong operations. In addition, the delay problem of task execution also affects the user experience. For scenarios that require immediate feedback, the waiting time of 2 minutes is too long.   Judging from the summary of opinions on the entire network, about 65% of the voices tend to support that QClaw is worth trying by ordinary people. This ratio reflects the market’s expectations for AI Agent tools and the trust brought by Tencent’s brand endorsement.

• 8. Industry evaluation (1667 chars)

  Reviews of QClaw in the technology media are generally positive, but generally point out that the product still needs improvement. 36 Krypton believes that QClaw’s performance on simple tasks is acceptable, and there are occasional surprises on tasks of medium complexity, but “domestic adaptation is necessary.” An analysis article by Tencent Cloud Developer Community pointed out that the secret of QClaw's success "is not the most advanced technology, but the best understanding of the needs of ordinary users."   In the comparison of competing products, QClaw has formed differentiated competition with products such as OpenClaw, Manus, and AutoGPT. As an open source framework, OpenClaw has strong technical capabilities but high deployment threshold, and is mainly aimed at the developer community. Manus emphasizes general intelligence and self-evolution capabilities, but is costly and lacks stability. In contrast, QClaw has chosen a "pragmatic" route: it does not pursue illusory general intelligence and only solves practical problems.   Industry analysts believe that competition in the AI ​​Agent track has shifted from technical competition to product competition and ecological competition. Technology leadership can only bring temporary advantages, while product experience and ecological construction are the decisive competitiveness. QClaw is well-positioned in this transformation with its WeChat entrance, zero-threshold deployment, and more than 5,000 community skills.   Tencent's strategic intentions have also been recognized by the industry. Through deep integration with the WeChat ecosystem, QClaw not only improves user stickiness, but also seizes the opportunity for Tencent in the local AI assistant track. This layout forms a direct competition with Alibaba Cloud, ByteDance, Baidu and other manufacturers, but the differentiated product positioning allows QClaw to find its own position in the ordinary user market.

• 9. Controversies and Risks (1420 chars)

  The main disputes faced by QClaw focus on product maturity and security responsibility boundaries.   In terms of product maturity, although the official iteration speed is relatively fast, problems such as task execution failures, response delays, and understanding deviations still frequently appear in user feedback. Some reviews pointed out that QClaw "is currently more like a running demo" and is still far from being a "usable product". This "semi-finished product" feeling may affect the long-term retention of users.   Security issues are a common challenge for AI Agent products. QClaw has high-privilege capabilities such as reading and writing local files, executing commands, and calling browsers. Once used maliciously, it may cause data leakage or system damage. Tencent’s response plan is a multi-layered defense system, including ClawScan threat intelligence scanning at the cloud prevention layer, sandboxed operation at the local isolation layer, sensitive call visualization at the privacy monitoring layer, and Token telemetry and log analysis at the behavioral audit layer.   However, there is always a tension between security and convenience. Security restrictions that are too strict may reduce the automation capabilities of the product, while permission settings that are too loose may bring risks. QClaw needs to find a balance between the two.   Another potential risk is the sustainability of the business model. The current daily computing power subsidy of 40 million Tokens is a considerable cost. It is unknown whether the long-term free strategy is sustainable, and whether users are willing to pay after switching to a paid model in the future.

• 10. Applicable groups and usage suggestions (979 chars)

  QClaw is best suited for people who:   Internet practitioners are one of the core target users. For practitioners who need to frequently process documents, data, and codes, QClaw's automation capabilities can significantly improve efficiency. In a remote office scenario, there is no need to bring a computer when going out, and emergency tasks can be handled through WeChat.   Content creators and self-media operators can also benefit from this. Work such as material collection, content organization, file archiving, and social media operation assistance can all be automated through QClaw, allowing creators to devote more energy to the creativity itself.   Freelancers and small and medium-sized business owners can use QClaw to manage clients, handle schedules, and generate documents. The scheduled task function can remind you of important matters, and the remote file sending function solves the need for file retrieval when you are out.   Students and researchers can use QClaw to organize data, read papers, and sort out knowledge. The automatic retrieval and review writing functions of academic papers are of auxiliary value for academic research.

• 11. Not recommended for use (633 chars)

  For industries such as finance and medical that have extremely high data security requirements, although QClaw adopts local deployment, risks still need to be carefully assessed. Operations involving sensitive information are recommended to be conducted in a completely isolated environment.   For scenarios with extremely high real-time requirements, such as high-frequency trading, real-time monitoring, etc., QClaw's current latency level may not be able to meet the needs.   If you are already a skilled developer and can deploy and configure OpenClaw yourself, then the "out-of-the-box" advantages of QClaw may not be outstanding to you. Instead, the flexibility and customizability of an open source framework may better suit your needs.

• 12. Summarize (1160 chars)

  QClaw represents an important step in the evolution of AI Agent from "geek toy" to "public tool". It may not be the most technologically advanced AI Agent, but it may well be the easiest for the average user to get started with. The three major features of WeChat entrance, local deployment, and zero-threshold configuration allow QClaw to find a unique ecological niche in the "Battle of 100 Shrimps".   The product is still in a rapid iteration period, and the release of the V2 version marks a move from "usable" to "easy to use." Upgrades such as multi-agent collaboration, application connectors, and Lobster Butler security functions demonstrate Tencent’s continued investment in products.   For the average user, QClaw is worth a try. The daily free quota of 40 million Tokens is enough to explore the various possibilities of the product, and the process of “shrimp farming” itself is also an interesting AI interactive experience. Of course, you need to be mentally prepared for the "semi-finished" state of the product and give it some time to grow.   For industry observers, the significance of QClaw is to verify a feasible path for the productization of AI Agent: not pursuing the technical brilliance of general intelligence, but focusing on pragmatic solutions to real needs. Whether this route can win in the long-term ecological competition deserves continued attention.

No in-depth report yet

• 

  ElizabethHarris_2020

  —

  The meeting minutes scene is really punchable. Our 88-minute regular department meeting used to take 75 minutes with third-party transcription and manual sorting, and we often missed the temporary tasks assigned verbally by leaders. QClaw directly reads the audio of local corporate micro-meetings without leaking the entire process. It outputs structured information after noise reduction, and produces minutes in 26 minutes. It automatically breaks down 6 fixed-job tasks and 3 high-level special requirements, and the division of labor directly binds the people and deadlines. The transcription of colleagues with strong dialects will drop to 72%, and multiple overlapping speeches will be merged into one paragraph, requiring manual fine-tuning of the attribution, but overall it saves most of the time.

---

• 

  cOSMWASMwIZ489

  —

  The morning broadcast is delicious. I set up a scheduled task at 8 o'clock every morning. QClaw adjusts weather skills, catches technology news, reads the first meeting in the calendar, and packages it into a paragraph to send to WeChat. This kind of "multi-skill linkage plus active push" is what distinguishes it from ordinary chatbots. You don't need to ask it to move. It's like having an extra butler who will take the initiative to worry about it.

---

• 

  Douglas.TaylorZ

  —

  I am really good at document proofreading. Our 4,200-word implementation plan took 110 minutes to manually check the proper nouns and formats word for word, and it was always returned by the leader to change the details. I built an exemption thesaurus in the skill center to lock the project codenames. The QClaw Level 4 proofreading produced a revised and final version in 21 minutes. I corrected 17 half-width punctuation marks, 8 codename capitalization errors, and marked 3 external ambiguities. The borders of documents with complex tables will occasionally be messy and require manual adjustment. There are zero mistakes in pure text, and it can adapt to 90% of ordinary documents.

---

• 

  rm2cr15bstrp

  —

  If you need to use files on your computer when you are outside, you can send a message on WeChat, "Send me the pdf on your desktop." It can really find and generate links to push back to the mini program. In the past, this kind of thing had to be connected remotely and then spent a long time, but now it is received while drinking coffee. This kind of experience of "being able to work forward without being in front of the computer" is really irresistible.

---

• 

  NancyJenkinsII

  —

  I asked it to go to Xiaohongshu to capture discussions about itself and make a report, but the result was that it was stuck at QR code login - it knew to scan it, but it couldn't pull up the browser and presented the login page to me, and just reminded me to scan the QR code in WeChat. The original OpenClaw can directly launch Chrome operations, but QClaw is awkward in this regard.

---

• 

  狗狗145

  —

  If you send the same command repeatedly, it will loop endlessly and keep returning the same message, so you have to stop it manually. This bug was quite discouraging in the early days, and I hope to add a limit on the number of retries to automatically stop it.

---

• 

  Carol.Perez007724

  —

  It is not possible to send pictures for analysis yet. It is a pure text interaction, and the multi-modal aspect is indeed lagging behind its peers. Sometimes I want to take a screenshot and ask it "What's going on with this pop-up window?", but I can only type a description myself first, which is quite awkward.

---

• 

  ShirleyEdwardsQ055

  —

  The 40 million tokens given out for free every day during the internal beta test are really impressive. If you use it heavily, you won’t be able to burn them all in a day. It saves a lot compared to connecting to the API yourself. However, complex task chains are still prone to overturning. Once I asked it to "remind me to eat in 1 minute", but the result was almost 1 minute late due to misalignment of the time benchmark. I asked it to write a tweet, but couldn't find any information, so I made up an irrelevant article regardless of the context. You still need someone to tell you everything about complicated things, so don't trust them all.

---

• 

  JGonzales_66

  —

  The overseas version is now available in Canada, Singapore and the United States. You can use WhatsApp and Telegram to send messages to your home computer. It can also file taxes, make daily fitness plans, and Up can help you run social media. Peter Steinberger, the founder of OpenClaw, reposted and praised it, saying that he helped repair the harness performance of the open source project and gave back the performance. This wave of overseas expansion has given him enough recognition.

---

• 

  Lambda_LPRoberts

  —

  The installation is really brainless. Double-click to download, wait 20 seconds, scan the WeChat code to bind, and start working in three minutes. Zero code and zero environment, my mother can use it.

---

• 

  realMalathyGugale

  —

  This is what scares me most about running it locally. Files and audio are not uploaded to the cloud, and I dare to use it to organize confidential meetings. Compared with those record-keeping tools that have to send recordings out, it is much more secure.

---

• 

  AndrewRoberts_2021

  —

  The multi-Agent collaboration in the V2 version is really great! I opened three agents at the same time to handle the reports, which doubled the efficiency. In the past, a task took more than ten minutes to run, but now it can be completed in a few minutes through parallel processing.

---

• 

  Susan.Barnes520

  —

  After two weeks of use, the 40 million tokens every day cannot be used up at all. A basic conversation only costs more than 10,000 tokens, and complex tasks only cost hundreds of thousands. Calculating, you can run hundreds of tasks a day, and the free quota is really generous.

---

• 

  Mia5r5

  —

  To be honest, the installation experience is perfect, but it is still a bit rough to use. I asked it to send a WeChat message for me, but I struggled for a long time without success. It feels more like a running demo now, and there is still a long way to go before it can actually be used.

---

• 

  DJackson_2022

  —

  The WeChat remote control function is perfect for me! I often travel on business, and suddenly I need to find a file or change a document. I just send a message directly on WeChat without having to carry a computer around.

---

• 

  CLtay

  —

  Local deployment must be praised. Data does not leave the computer, which is very important for an industry like ours that has high privacy requirements. No matter how good the AI ​​tools in the cloud are, data security is always a concern.

---

• 

  琉璃897

  —

  Response speed is indeed an issue. I asked it to help me organize desktop files, and it took almost two minutes for me to reply. If it's an emergency, this delay can be really worrying. Hope it can be optimized later.

---

• 

  Ashley527

  —

  The concept of shrimp farming is quite interesting. After using it for a while, I found that it really started to understand my habits. For example, if I am used to writing emails more formally, the tone of the content it generates will also be more formal.

---

• 

  TPatel369700

  —

  The more than 5,000 skills in the skills market are so abundant! From weather query to GitHub operations, basically everything you need for daily office work is covered. You can also create your own skills and share them, and the ecosystem is doing well.

---

• 

  吕强哲

  —

  In actual testing, I asked it to help me write a note in Xiaohongshu, but it got stuck on the login QR code and could not be moved. The stability of third-party skills really needs to be improved. Many scenes look good but have various problems in actual use.

---

• 

  墨染84

  —

  As an ordinary user who doesn’t understand coding, QClaw is the easiest AI agent I have ever used. I tried OpenClaw before, but was immediately discouraged from the step of configuring environment variables.

---

• 

  NatalieCooperZ45

  —

  I asked it to help me process the Excel table. As a result, the data from the two regions were merged incorrectly, and the column names were also recognized reversely. The PPT chart and text generated in the end did not match at all, and I had to change it manually.

---

• 

  Maria_MartinK

  —

  The function of Lobster Butler is quite practical. Once it almost deleted an important file on my desktop. Fortunately, it safely intercepted the pop-up reminder in time, otherwise it would have been a tragedy.

---

• 

  Roger_VasquezJr

  —

  The WeChat entrance is indeed a stroke of genius. There is no need to download a new app or learn a new interface. You can control the computer just by sending messages on WeChat. This design is very in line with the usage habits of ordinary users.

---

• 

  5h1ae2m94_n

  —

  It's really annoying that multitasking is not supported in parallel. When executing a new task, existing tasks will be interrupted. Sometimes a task will run for half a day, and you can't even check anything in the middle.

---

• 

  Douglas839_m

  —

  The installation process of ClawBot is too fiddly. The official said that it can be opened with one click, but in fact it required installing several software packages. The permission denied prompts popped up repeatedly, and it took nearly an hour to install them reluctantly.

---

• 

  8kk5uebg

  —

  After comparing similar products, QClaw’s pricing strategy is indeed conscientious. The client is free, and 40 million tokens are given out every day, which can basically meet daily office needs, unlike some products that charge at every turn.

---

• 

  AGbar

  —

  I asked it to help me summarize the AI ​​news in the last 24 hours, but it took more than 10 minutes. The timeline was still confusing and the content was not comprehensive. The effect is not as good as if I search manually or ask using ChatGPT.

---

• 

  NathanRuiz520

  —

  The model switching function is very useful. Several built-in domestic models have their own characteristics. Kimi is used for processing Chinese tasks and DeepSeek is used for code generation. You can choose flexibly according to your needs.

---

• 

  JMorgan_637

  —

  There are a lot of restrictions in WeChat. QClaw is an "application" and not a "contact". It cannot be added to a group, nor can it be forwarded directly for processing. You have to copy and paste it every time, which is quite troublesome.

No reviews yet