Open Claw
Overview
I set up an Open Claw agent called Anees, which is my personal AI assistant. Rather than giving it access to my own accounts, I gave Anees his own Gmail and Linear accounts. I also gave him his own device (currently running on a MacBook Air M2). I can message him on Telegram via my phone and he handles my inbox triage, surfaces deadlines from Linear and my calendar, drafts replies, and logs tasks.
The problem
After seeing a lot of hype online about building personal AI assistants using Open Claw, I wanted to try it for myself, primarily to prove I could actually set something like this up, but also to get the real benefits of having an assistant that is always on and does not need hand-holding. I had watched videos and read plenty of guides, and eventually I just wanted to start. I assumed I would need a Mac Mini to run it properly, but I ended up using an old laptop I already had, and decided to spend the long Easter bank holiday weekend figuring it out. I had never set up anything like this before. There was a fair amount of fiddling with API keys and terminal commands along the way, but that felt like a reasonable price to pay both for what I would learn and for what I would get on the other side: an agent that handles the menial tasks, the things that would otherwise slip through the cracks.
The solution
After wrestling with the terminal, I managed to install Claude using my ChatGPT subscription. I opted to use Telegram because it was easier to set up programmatically compared to WhatsApp. I spent considerable time editing markdown files (AGENTS.md, TOOLS.md, SOUL.md, etc) to give Anees a distinct personality and clear instructions on how to use the tools I connected to him. The setup was time-consuming, but I'm relatively happy with the results. I did run into a few issues along the way. Primarily my MacBook sleeping when left unattended, which meant Anees would go silent. I solved that by downloading Amphetamine, an app that keeps the MacBook awake even when the screen closes. There are still iterations I'd like to make, and I plan to spend more time refining the system and eventually spin up a couple of other agents to handle finance and media-related tasks. Google has also released Gemma 4 which means I could swap from ChatGPT and end up running my agent free of cost! Overall, the learning experience has been genuinely rewarding, and I'm excited to build on it next.
