Whatsapp is the biggest pain in the arse. Also Codex is a lot cheaper than CC. And you can use it with your ChatGPT Plus/Pro subscription, which gives you ~$200 of API credit equivalent for $20/mo.
Telegram was so easy, maybe you can explain why you’d need to go through the rigmarole of using WhatsApp. Telegram is now my OpenClaw (fka Clawdbot) hub.
Thanks for the response. And the article! Have you posted your favorite use cases for OpenClaw yet?
Telegram was also pretty accessible for my wife. I added her to a group chat with my OpenClaw bot so we could send through our receipts of our purchases and automatically track our spending.
Also, not as accessible, but I’m using a DeepSeek API key because it’s the cheapest option I could find for a pretty smart model, and it’s mainly just managing databases on my Mac mini for tracking all of the things.
Great practical guide, Aman! The hardware and infrastructure decisions you cover are exactly what most tutorials skip over. I went through similar debates when setting up my own system - the Mac Mini vs cloud trade-off is real, especially when you factor in the cumulative API costs over months of continuous operation.
Your point about dedicated hardware resonated with me. I ended up building Wiz, my own AI agent running on Claude Code, and the infrastructure decisions were surprisingly complex. The phone number situation alone took me longer than expected to figure out properly - eSIMs are definitely the way to go for this kind of setup.
One thing I found interesting in my own journey was analyzing the actual costs at scale. Clawdbot's popularity (44K+ stars now) means there's a lot of people running into the same infrastructure questions you're answering here. When I dug into the numbers for my setup, the token costs added up faster than I initially expected.
Whatsapp is the biggest pain in the arse. Also Codex is a lot cheaper than CC. And you can use it with your ChatGPT Plus/Pro subscription, which gives you ~$200 of API credit equivalent for $20/mo.
Are you getting good results with GPT+OpenClaw? Haven’t tried it yet
It’s good on 5.3. When my subscription usage runs out and I have to switch over to the API (which only has 5.2) it becomes quite frustrating.
This is so timely & thorough. Thank you for this post!
Telegram was so easy, maybe you can explain why you’d need to go through the rigmarole of using WhatsApp. Telegram is now my OpenClaw (fka Clawdbot) hub.
I’ll add this in - I prefer WhatsApp personally because it’s a product I already use
Thanks for the response. And the article! Have you posted your favorite use cases for OpenClaw yet?
Telegram was also pretty accessible for my wife. I added her to a group chat with my OpenClaw bot so we could send through our receipts of our purchases and automatically track our spending.
Also, not as accessible, but I’m using a DeepSeek API key because it’s the cheapest option I could find for a pretty smart model, and it’s mainly just managing databases on my Mac mini for tracking all of the things.
Thank you very much for sharing your insights!
Very useful instructions. I was able to get it up and running within a couple of hours. Thank you!!
Great practical guide, Aman! The hardware and infrastructure decisions you cover are exactly what most tutorials skip over. I went through similar debates when setting up my own system - the Mac Mini vs cloud trade-off is real, especially when you factor in the cumulative API costs over months of continuous operation.
Your point about dedicated hardware resonated with me. I ended up building Wiz, my own AI agent running on Claude Code, and the infrastructure decisions were surprisingly complex. The phone number situation alone took me longer than expected to figure out properly - eSIMs are definitely the way to go for this kind of setup.
One thing I found interesting in my own journey was analyzing the actual costs at scale. Clawdbot's popularity (44K+ stars now) means there's a lot of people running into the same infrastructure questions you're answering here. When I dug into the numbers for my setup, the token costs added up faster than I initially expected.
I wrote up my experience comparing Clawdbot's approach to building a custom agent, including the real costs I'm seeing - might be useful context for anyone deciding which path to take: https://thoughts.jock.pl/p/clawdbot-deep-dive-personal-ai-assistant-2026
"Whatever they decide to name it tomorrow" — the C&D that caused those names spawned a $16M crypto scam in the 10-second window. $2,600/yr subscriber watched Anthropic's side for a year: https://aiwithapexcom.substack.com/p/after-nearly-a-year-on-claude-max