OpenClaw: The AI Agent Everyone Is Talking About
OpenClaw is a free, open-source AI agent that connects LLMs to your actual software. It has 247k GitHub stars and a wild origin story involving trademark disputes and an OpenAI acquisition.
OpenClaw: The AI Agent Everyone Is Talking About
If you've been anywhere near the AI dev community in the last few months, you've heard of OpenClaw. With 247,000 GitHub stars and 47,700 forks as of early March 2026, it's the most popular open-source AI agent project by a wide margin. But the story of how it got here is almost as interesting as the tech itself.
The origin story
An Austrian developer named Peter Steinberger built the original version and published it in November 2025 under the name "Clawdbot." Anthropic sent a trademark complaint, so on January 27, 2026, Steinberger renamed it to "Moltbot." Three days later, he changed it again to "OpenClaw." On February 14, 2026, Steinberger joined OpenAI, and the project moved to an open-source foundation to keep it vendor-neutral.
That's three names in under a month, the creator joining a competing AI lab, and a governance transition to a foundation. Wild stuff for a project that's barely four months old.
What OpenClaw actually does
OpenClaw is a Node.js service that lets you connect any LLM to real software through messaging platforms. You talk to it over Signal, Telegram, Discord, WhatsApp, Slack, or iMessage, and it does things on your behalf.
It's model agnostic. You can point it at Claude, GPT-4o, Gemini, DeepSeek, Grok, or run something local through Ollama. The architecture has five components:
- Gateway routes incoming messages from your messaging platform to the right handler
- Brain runs a ReAct reasoning loop against whatever LLM you've configured
- Memory stores context as Markdown files on disk (simple, but it works)
- Skills are plugins that give the agent new capabilities
- Heartbeat is a scheduler for recurring tasks
What can it do?
Quite a lot. It can read and write files on your system, run shell commands, browse websites, send emails, and control external APIs through skills. The ClawHub marketplace has over 10,000 community-built skills covering everything from calendar management to database queries.
I've been using it for about six weeks and the thing that surprised me most is how natural the messaging interface feels. Instead of learning yet another CLI or web UI, you just text it like you'd text a coworker. "Check if the staging deploy finished" or "summarize the Jira tickets assigned to me this sprint." It figures out which skills to invoke and chains them together.
Where it falls short
The security model. OpenClaw gives the agent broad access to your system and trusts that skills from ClawHub are safe. They often aren't. I'll cover this in a separate post, but the short version is: audit everything you install and don't run it on a machine with access to anything you care about losing.
Also, Memory being just Markdown files means context can get stale or inconsistent in long-running setups. It works fine for personal use, but if you need reliable state management, you'll want to look at the database-backed memory plugins.
The bottom line
OpenClaw is genuinely useful. The messaging-first approach, model flexibility, and the huge skill ecosystem make it the easiest way to get an AI agent doing real work. Just go in with open eyes about what it can access on your system.
