Skip to main content

3. How OpenClaw Works

Action Step

Complete this before moving on.

Try three slash commands with your OpenClaw agent: /status to check your usage, /compact to compress context, and /context list to see what's loaded.

Comment in Slack

Post your answer in your onboarding channel.

When would you choose to use Claude Code over OpenClaw for a task, and why?

Training Guide

You just used OpenClaw. Now let me explain what was actually happening under the hood — and you're going to realize you already know most of it.

The Architecture (Simple Version)​

Your agent lives on a server in the cloud. That server runs a program called the "Gateway" — it connects Slack to the Claude API. When you send a message in Slack, the Gateway receives it, sends it to Claude, Claude responds, and the Gateway sends the response back to Slack.

It's a middleman between you and the same Claude you've been using.

Sessions = Threads​

  • In Claude Code: each VS Code chat tab = a session
  • In OpenClaw: each Slack thread = a session
  • Same concept, different interface
  • DMs with your agent = one continuous session (unless you reset)

Remember when we talked about context windows filling up? Same thing happens here.

Context Windows Still Apply​

Your agent still has a 200k token budget per conversation. When it fills up, compaction happens (auto-summarize) and nuances get lost. Same telephone game analogy from the token management training.

The difference: OpenClaw handles this more automatically than Claude Code. It saves important stuff to files BEFORE compacting (pre-compaction memory flush). It writes daily notes so it can look things up later. It's doing what you learned to do manually with handoff documents — but automatically.

Slash Commands You Already Know​

  • /reset or /new = fresh session (like /clear in Claude Code)
  • /status = see usage/quota
  • /model = switch AI models
  • /skill <name> = run a skill (same concept)
  • /stop = abort what it's doing
  • /compact = manually compress context (like manual compact in Claude Code)
  • /context list = see what's in the current context

You already know what these do. The commands just look slightly different.

What You Lose (The "Headless" Tradeoff)​

  • No file explorer — you can't see folders and files on the left
  • No diff view — you can't see green/red line-by-line changes
  • No visible token counter — you can't see how much context is left
  • No copy-path workflow — you can't right-click a file and paste the path

This is the tradeoff: you gain always-on access from your phone, but you lose the power-user cockpit.

What You Gain​

  • Always on — your agent runs 24/7, even when your laptop is closed
  • Phone access — message it from anywhere via Slack mobile
  • Persistent memory — it remembers you across days and weeks
  • Automation — cron jobs let it do things on a schedule

Claude Code is your workshop. OpenClaw is your assistant who's always in your pocket.

When to Use Which​

  • Use Claude Code when: you need to create documents, work with files, see diffs, do deep project work, need the full cockpit
  • Use OpenClaw when: quick question from your phone, need something while you're not at your desk, want automated tasks, simple Q&A, transcript processing