Self-hosted AI agent you run like a digital employee.
OpenClaw is a self-hosted AI agent you run on your own server rather than a hosted app you log into. It is free and open source under the MIT license, so the only real costs are the server it runs on and whichever AI model you connect to it.
Once it is set up, you can wire it into a channel like Telegram, so you message it directly and it works through tasks on its own, more like an employee than a piece of software you open.
Ian runs roughly a dozen OpenClaw agents from Telegram. In Episode 003, he set one up live on camera, starting from a blank server: hardened it, installed a model, and connected it to Telegram in under an hour. Pointed at the Episode 002 brief, an OpenClaw agent had already produced GotoBuild's full hosted brand kit. In Episode 005, he ran multiple OpenClaw agents in parallel to build Sunday Crumb, handling the branding, site copy, order flow, and admin view all at once.
We keep OpenClaw in the stack because it proves an agent can act like staff, not just answer a question once and stop working. The practical takeaway here isn't to go harden a server yourself. It's that a working AI employee is one blank server and about an hour away, and once it's running, you manage it the same way you'd message anyone else on your team.
OpenClaw itself is a technical tool: setting it up means provisioning a server, hardening it, and connecting a model and a channel like Telegram. That's real infrastructure work, not a beginner-friendly signup flow. It's worth knowing about anyway, because it's what proves an AI agent can run like an actual employee instead of a chat window. Most operators will have someone else (or, increasingly, another AI) do the setup, then talk to the finished agent like a coworker.
Genuine substitutes for OpenClaw that we've also run on stage:
Yes, OpenClaw has a free tier. Free and open source (MIT license). You pay only for your server and your model.
Free and open source (MIT license). You pay only for your server and your model.
Ian Kilpatrick used it in Episode 002 for producing GotoBuild's full hosted brand kit; Ian Kilpatrick used it in Episode 003 for setting one up live, from a blank server to a working Telegram bot; Ian Kilpatrick used it in Episode 005 for running multiple parallel agents for Sunday Crumb's branding, copy, orders, and admin view; Ian Kilpatrick used it in Session 006 for named agents (Atlas, Saint, Saucy) that built Echo Check, run the Saints Classical LMS, and give every BrandSauce employee their own agent; Ian Kilpatrick used it in Session 007 for the $18-a-month starter combo with DigitalOcean; behind his 10 apps in 10 weeks, the school LMS, and the AG4NTS build.
OpenClaw itself is a technical tool: setting it up means provisioning a server, hardening it, and connecting a model and a channel like Telegram. That's real infrastructure work, not a beginner-friendly signup flow. It's worth knowing about anyway, because it's what proves an AI agent can run like an actual employee instead of a chat window. Most operators will have someone else (or, increasingly, another AI) do the setup, then talk to the finished agent like a coworker.
Tools we've also used live that can stand in for OpenClaw: Hermes.
| Category | AI agents & automation |
|---|---|
| Pricing | Free and open source (MIT license). You pay only for your server and your model. |
| Free tier | Yes |
| Used live in | Episode 002 and Episode 003 and Episode 005 and Session 006 and Session 007 |
| Operator(s) | Ian Kilpatrick |
| Website | openclaw.ai |
On AI4NTP, every tool is shown live by a real operator. These are the people who demoed OpenClaw on stage.
Explore other tools we've demoed live at AI4NTP:
See tools like OpenClaw set up live, step by step, on AI4NTP, or get hands-on help putting it to work.