"How to make your silly Openclaw assistant 10 times smarter?" 1. Upgrade the "brain": Choose a model with stronger reasoning capabilities. Claude 3.5 Sonnet is currently the most recommended. If you're looking for a domestic alternative, use minimax, which costs only a few hundred RMB for an annual subscription. If you have a high-end graphics card, you can run Llama 3.3 or DeepSeek-V3 locally through Ollama for completely private local inference. 2. Deeply feed background information. Tell your assistant your background information, investment logic, thinking framework, personal habits, etc., to establish "long-term memory"—this is Openclaw's advantage (this information exists in *.md files and can also be modified directly). 3. Enable proactive mode. Proactivity is another advantage of Openclaw. For example, you can tell it to check the Twitter KOLs in the list every hour and summarize the information for you. 4. Install professional skill plugins. You can see the default installed skills in the Openclaw skills list, but some require pre-installed modules to activate. You can have Openclaw help activate them, but be sure to pay attention to security; there may be malicious code in the skills. Trusting officially certified ones is safer, but you should also double-check. Additionally, you can install non-certified skills, but this carries a higher risk, so you must verify repeatedly before installation. 5. The highest level—let it evolve by itself (Self-Improving Loop). I haven't tried this myself, but those who are bold can give it a shot 😂. You can talk to Openclaw about "how to enable self-evolution mode," which can turn your Mac Mini into a complex AI matrix. Do you have any other tips?

Supplement 1. There is a typo in the article. The best version of Claude is Claude 4.5 Sonnet. 2. The key for Claude premium members is also usable, which is definitely better than directly consuming tokens, but the consumption speed of the member key is also relatively fast.