Vendor-neutral · Beginner · 3 lessons · about 19 minutes · updated 2026-09-13
OpenClaw 101: The Free Agent Gateway With a Hidden Token Bill
OpenClaw is a free, open-source bridge between Discord/Slack/Telegram and an LLM agent. Three lessons on what it is, what it actually costs once it's running, and setting up your first bot.
For anyone who saw "OpenClaw" mentioned as a way to get an AI agent into Discord or Slack and wants to know what it actually is, whether it's really free, and how to try it. No prior coding required, though setup uses a terminal.
Course outline
- 1What OpenClaw is, and why the name changed twiceOpenClaw is an open-source "agent gateway": it connects a messaging platform to an LLM agent so the bot in your Discord or Slack server can actually do things, not just chat. It's free software — the model behind it isn't.6 min · 3 objectives
- 2The real cost: whichever model you plug inOpenClaw's own price is $0. Its running cost is entirely the token bill of the model behind it — and that bill can vary by more than 10x depending on which model you choose.7 min · 3 objectives
- 3Setting up your first OpenClaw botA recorded terminal walkthrough of the shape a first OpenClaw setup takes: cloning the project, adding a model API key, and starting the gateway.6 min · 3 objectives
Frequently asked
What is OpenClaw?
An open-source, MIT-licensed agent gateway connecting messaging platforms (Discord, Slack, Telegram) to an LLM agent loop, so an AI bot can read and act in your server rather than answer in a single chat window.
Is OpenClaw free?
The software itself is free and open-source with no license fee. Running it isn't free, though — every message it processes is billed as a normal API call to whichever model you connect it to.
How much does it cost to run an OpenClaw bot?
It depends entirely on the model backend and message volume — this site's worked example puts a busy bot anywhere from about $0.10 to $1.25 per 1,000 replies depending on which model is plugged in, or near-zero-per-token if you self-host an open-weight model instead.