Back
Post
Who is building with Clawdbot (now called Moltbot)? 🦞
Just to share some interest insights and discuss good practices, tips and tricks.
Here are my findings so far:
Here are my findings so far:
- I had to spin up an EC2 instance on AWS because we're too broke to buy a Mac Mini.
- I've had to manually configure another 9084098534 things from the terminal. It's not as plug & play as they say.
- Running it from a private IP in Antigravity's terminal has made it easier when I get stuck and more secure.
- I've been using chatgpt-4o-mini and I've only spent about $2.6 USD in approximately 10 intense hours of back and forth. However, in the short term I want to move to a local open-source model.
- Creating an email for it to manage has been a pain in the ass—the typical "Gmail" or "Hotmail" have thousands of blocks, and rightfully so, so I think it'll stay without email for now.
- Cron jobs are super useful, but I think for that I'd rather just have a dedicated instance to run the jobs.
- 90% of the things I've had to do so far have required an intermediate technical level—I still wouldn't recommend it to someone who doesn't use the terminal.
- I love Anthropic a lot but they're looking pretty jealous with the restrictions and bans, but oh well... some will argue otherwise.
- Apparently the best option for connecting to my desktop would be Tailscale, but I'm still wrestling with a couple of concepts that change because the instance is deployed on a private IP.
- So far I haven't connected to external sites other than with my own data and model outputs. We'll see how prompt injection goes.
- I still haven't started refining the agent's .md files, but that's also in the backlog.
And I could keep going...
Did I know all this when I started the project? Definitely not.
I think what I'm taking away more than "achieving the goal of having a 100% automated AI assistant" is learning a ton of things I didn't know I didn't know and that's invaluable.
👋 Join WIP to participate
it's very tecnical and the immediate setup was a bit involved, took me 2-3h but it configures MCP integrations on its own. had a bit of trouble with getting my morning, lunch and afternoon brief properly set up. not all are sent out.
I went the tailscale route, VPS is running a gateway and my macbook is a node. I sync obsidian via syncthing.
you can check my profile for stuff I've configured it for, it's really awesome. prompt injection is a point tho, and having that much access to data as well.
It's OpenClaw now 🤣
I installed mine the first weekend it came out. It took ~15 minutes, and then I took my kids to the library to pick out books. I sat in a chair while they were looking and had it bookmark and summarize a few links. For the hell of it, I asked it to take a screenshot of a few websites, and it sent them back 30 seconds later, to my amazement.
I haven't connected email to it, but I plan to connect my calendar so that it can build out a better daily briefing for me and possibly schedule a few things for me.
I'm running it on a Mac Mini (before it was cool), but i wouldn't buy a new machine for it or run it on a VPC unless you need it to. Run it on Tailscale, with a firewall blocking everything.
I think the platform's power lies in cron and memory. Its ability to just do things is killer too. I was in bed and got a downtime alert from some server issues I hadn't sorted out. I asked it to "ssh node..." and reboot it, and a minute later, my server was back up.
So I'm a fan, but I haven't had it do anything that generates revenue. Yet.
4-6 hours to get it up and running. But I also started from a fresh Mac install, and had to figure out how to install it in a VM (used UTM), so I added a couple of hoops. It has a ton of hardware (4 CPUs and 16GB of RAM). I want to figure out how to install a local LLM and do a quality comparison on coding tasks.
One of the first tasks I gave it to do, it was like, ok, I've done it, and now here's a bunch of work for you to do. I had to send it back to figure out how to do it because I didn't want more tasks for me.