Look, there's no denying hype, people are buying MacBooks left and right (they don't really have to), and it's really amazing. It's also very brittle, prone to issues and bugs. It doesn't take away from how amazing it is.

I figured I'd add a few tips and tricks (and keep adding as I go along as replies to this article) - mainly for me, but also in case anyone else runs into the same things as me. Here goes:

🦞OpenClaw tips and tricks

1. Models - Choose Opus 4.5 unless you absolutely can't afford it
   Opus 4.5 is the most incredible model, it's worth getting the Max subscription for your Clawd, but if that's too much, don't use the API directly, you can get free Opus 4.5 via the Google Antigravity connector. Some good alternatives for API keys are MiniMax M2 and GLM 4.7.

2. Use Sub Agents but be careful of tokens
   You Clawdbot can spin off subagents to do a bunch of tasks, not only does this speed up parallel execution, it also frees up your Clawed to respond back + lowers it's context window. You can even specify which model you want per what agent. For context heavy stuff, use Gemini Pro (1M context) agents and send only summaries to 🦞 Be mindful of tokens if you're using via API!

3. You don't have to buy a new Mac Mini for your Clawdbot (but it's cool and I couldn't resist)
   The gateway (the brain if you will) of Clawdbot can run even on a $100 RaspberryPi according to the creator. You can connect many nodes like your Macbook securely with Tailscale (which is free and very easy to setup). But if you do buy a new mac mini, your Clawdbot will have it's own home!

4. Don't ask it to write code - have it plan and supervise
   Have an instruction that Clawd should always drive another agent like Claude Code/Codex to write the code it needs written. This frees your clawdbot to monitor and be helpful, and reduces the context window significantly.

5. ⚡ Capital instructions symbol
   Teach your clawdbot that when you specify (via emoji or text) a capital instruction, it should save it to it's memory first, then execute. This is a great way to make your correction instruction really stick for next sessions.

6. Connect it multiple channels
   Telegram is the easiest to set up (literally a DM to

   @botfather

   on TG and you're good) but WA is also great. Clawdbot understands which channel it's being spoken from. Protip: Adding Clawdbot to a groupchat with your Family can get all kinds of fun.

7. Give it access as much as you feel comfortable
   A lot of the magic comes from giving Clawdbot access to a wide array of information and data about you, files, emails, todos etc. As far as you feel comfortable, give it an ability to help you with your daily life. But be careful with mission criticial emails and accounts. Passwords via a 1Password service account is a great way to limit it's scope and keep it secure

8. Get comfortable with Clawdbot CLI
   Especially in the beginning and setup, clawdbot command line interface is where you would set up, debug and update configs. It helps learning it just a little bit to be able to restart if issues happen. They will happen.

9. Ask Clawdbot to learn stuff permanently and often
   This is what everyone is super excited about, Clawdbot has incredible self improving abilities, you can chat with it for 3 hours, then ask it "what from the above should you learn forever, write a skill" and it will KNOW it next session. This is the richest surface area of using Clawdbot and why all of us are super excited

10. Give it Grok API access
    Grok 4.1 is the best real time researcher about information you can get access to because it has access to web search + X search. Sign up for their API, and tell Clawd that everytime you want research, it should use a Grok skill. This is cheap and much better than the brave API suggested by default.