great! The flowers are 899 and 999, and after a few days, I couldn't afford to raise them. Even after unloading, I still need 299 chives, which is how they were harvested. To be honest, I think the popularity of crayfish is a good thing, indicating that there is a real demand for AI to be able to "work". Previously, AI was just a chat companion, but now AI can help you book flights, check data, and write code, which is the true productivity tool. But one thing to say is that the threshold for raising lobsters is still a bit high at this stage. Not everyone knows how to configure environments, tune APIs, and work on cloud servers. What does the scene of the long queue at Tencent's entrance indicate? There are many people who want to eat this hot meal, but very few can actually take care of it themselves. I have at least seven or eight friends around me who have tinkered with OpenClaw and eventually abandoned it, and the reasons are quite consistent. Many people had high expectations for it at first, thinking that they could finally build their own personal productivity system. However, after getting started, they found that the problems with the bare version were too fatal, and they gave up soon after. For example, if you climb any task, your token can burn nearly 20 dollars. Many friends have also roast that the native OpenClaw is more like a toy that can be used to run Demos. A little bit into the real scene will start to drop the chain. Many people either end up spending a lot of money to hire engineers for custom development, or simply give up. I have been tinkering around for half a month without writing a single line of code. With just three ready-made tools, I have solved most of the most challenging issues in the native OpenClaw experience. Sharing this method today should be helpful to many people. The reason why OpenClaw has become the first choice for many people to build personal agents is simply because of its low threshold, fast learning speed, and decent ecosystem. But the native version has three obvious shortcomings that basically block most people who want to really use it. Three core issues of native OpenClaw 1. There is no communication channel specifically designed for AI Once AI sends messages in bulk or frequently sends content, it is easy for the platform to mistake it for abnormal behavior, ranging from limiting traffic to directly banning accounts. In such cases, it cannot be integrated into the real workflow. 2. Poor memory retrieval efficiency The native solution uses SQLite with keyword search. With a large amount of content, the search speed will become slower and often inaccurate. After a prolonged period of time, the more memory accumulates, and the consumption of tokens will also skyrocket. 3. Special fee token for web browsing Many automated browser tools will stuff the entire page of HTML into the model when working, and a few web pages can fill the contextual window. In this way, not only is the experience poor, but the cost is also outrageously high, and using it a few times will make people lose patience. The three tools I added later 1. Agent Mail This is a product developed by a company invested by YC, which is essentially a dedicated email for AI. Up to three free independent email accounts can be opened. Its advantage is that it is not easily treated as an abnormal account by the platform. Whether used for automatic processing of personal emails, forwarding news summaries, or as the first layer of filtering for customer service email, it is very useful and can be installed directly through skill files or ClawHub. 2. QMD This is an open-source memory tool developed by Toby L ü tke, CEO of Shopify. It replaces the native search method that relies solely on keyword matching and uses local mixed search, which includes three layers of capabilities: keyword matching, vector search, and local small model rearrangement. The entire process runs locally without additional costs. Even if you can't remember the exact keywords, you are likely to be able to retrieve the historical content, and the experience will be much more stable than the native solution. The problem of increasing difficulty in using memory will also be alleviated. 3. Agent Browser This is a command-line browser developed by Vercel. Compared to common Playwright solutions, its biggest advantage is the cost savings of tokens, with official claims of up to 93% savings. It can click on page elements, take screenshots, fill out forms like a real person, and has built-in prompt injection protection, so accessing third-party websites is relatively safer. A more practical point is that it can even operate desktop applications based on Electron such as Slack, Notion, and VSCode. Many services that do not have open APIs can still be directly used for interaction. The method of integrating OpenClaw is not complicated either Integrating these three tools into OpenClaw eliminates the need for you to manually navigate through documents or manually configure them. My own approach is to use Claude Code directly, which can also avoid the expensive API cost of OpenClaw itself. The entire process actually consists of three steps. The first step is to prepare the environment Install the dependency packages that need to be installed first, such as the npm environment required by ClawHub and the local binary files corresponding to Agent Browser. These installation packages are officially provided, just follow the instructions. Step two, feed the official technical documentation of the three tools directly to Claude Code You don't need to organize the key points yourself here, just throw in the original document directly. Step three, have Claude Code generate the corresponding skill.md and configuration Give Claude Code a simple command to generate the corresponding skill.md file and configuration for you, and then automatically mount it into the OpenClaw directory. If the whole process runs smoothly, it can be completed in about an hour. In addition, it is best to entrust the installation and configuration of this development oriented work to Claude Code, rather than letting OpenClaw do it on its own. Because this type of task itself consumes a lot of tokens, if left to OpenClaw to handle, the cost will burn quickly, and many of the expenses are unnecessary. A more cost-effective way is to have Claude Code set up the environment and capabilities, and then have OpenClaw run the configured features. By dividing the work in this way, the cost will be much lower. Actually, many Agent frameworks are about to be implemented now, so there's no need to immediately think about building your own wheels or writing a bunch of custom code. There are already many mature tools on the market that can precisely address these common pain points. As long as the tools are selected correctly and combined in the right way, without spending too much time and cost, experimental tools that could only be used for demonstration can be turned into efficiency tools that can truly work for you.