Everyone's been talking about OpenClaw, so I finally sat down and dissected how it actually works under the hood.
OpenClaw is the open-source AI agent that went from 9k to 145k+ GitHub stars in days, the fastest repo to hit 100k in GitHub history! The architecture is surprisingly simple. So I made this diagram to break it down:
- You send a WhatsApp message (or Telegram, Slack, Discord ; it doesn't care) : "Make a reservation at Nobu for 2 at 6 PM on Friday".
- A library called Baileys picks up your message via the WhatsApp Web protocol and passes it to the OpenClaw Agent, which is running locally on YOUR laptop.
- The Agent asks an LLM (Claude, GPT, DeepSeek, Llama; your choice) "What do I do?"
- The LLM comes back with a plan: "Try OpenTable. If that doesn't work, call the restaurant directly."
- The Agent executes that plan using its local tools: browser automation, shell commands, file access, even voice calls via Vapi.
- Confirmation comes back to you on WhatsApp: "Done! Table booked for 2 at Nobu, Fri, 6 PM" The key insight: the orchestration stays on your machine. Your files, memory, and logs never leave. But messages travel through WhatsApp's servers, and the LLM call goes to the cloud (unless you run a local model). But here's what really sets it apart: OpenClaw isn't just reactive. It's proactive. It has a "Heartbeat" system. Every 30 minutes, it wakes up, checks a task list you define, and decides if something needs your attention. It can scan your inbox, flag what's urgent, check your calendar for conflicts, and message YOU first. You can also set up cron jobs for scheduled tasks: morning briefings, daily email triage, weekly report summaries. You set it once and it runs forever!
This isn't a chatbot you talk to. It's an agent that thinks ahead for you.
What's the first task you'd hand off to a proactive AI agent?