The OpenClaw Video Course
Watch all 16 lessons of the OpenClaw bootcamp in one continuous 4 hours 20 minutes video, with hard chapter markers at the start of every day so you can jump anywhere. The OpenClaw video course is free, hosted on YouTube, and embedded directly on this page so you can watch without leaving the site.
Standard YouTube controls work, including 2x speed, captions, and chapter skips. Every chapter also has a long-form written companion page if you would rather read than watch.
Trouble loading? Watch the full course directly on YouTube: https://youtu.be/lyZSFSe3Gj0
Don't have 4 hours? Pick a watch path.
The video is structured so you can stop after any day and still have shipped something useful. Four recommended watch paths:
30-min minimum viable watch
~30 minMental model + install. By the end you understand what OpenClaw is and have it running on your machine.
Days: 1, 2
1-hour foundation watch
~60 minMental model, install, channels, SOUL.md. By the end you have an agent talking back on Telegram in your voice.
Days: 1, 2, 5, 7
90-min build-something-real watch
~90 minFoundation plus HEARTBEAT.md. By the end you have a proactive agent that runs morning briefings on its own.
Days: 1, 2, 5, 6, 7
Full 4-hour watch
~260 minAll 16 lessons. By the end you have a production OpenClaw agent on a VPS, optionally multi-agent.
Days: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
Chapter list with YouTube timestamps
Sixteen chapters in the OpenClaw video course. Each timestamp deep-links into the YouTube video at the exact second the chapter starts. Each chapter also has a written companion page.
The mental model. OpenClaw is a persistent agent daemon, not a chatbot. Reactive vs proactive AI, the gateway, channels, memory, soul, and tools.
Install Node, run the curl installer, walk the onboard wizard, get the gateway running on port 18789, open the dashboard, connect Telegram via BotFather.
Why OpenClaw is model-agnostic. Anthropic Claude, OpenAI GPT, Google Gemini, Chinese models (DeepSeek, Kimi K2, GLM 5). Run Llama 3.2 fully local with Ollama.
Two-tier processing (cheap model for HEARTBEAT, premium for interactive), prompt-bloat audits, prompt caching, spending caps. Drop $70/mo to ~$17 with the same agent.
Wire your agent into the apps you already use. Per-channel auth, allowlists, formatting differences, voice notes, and the cross-channel memory test.
Stop being reactive. Write HEARTBEAT.md so your agent runs morning briefings, deadline checks, and system monitors on a schedule, with the heartbeat:ok silence guard.
Replace the generic chatbot voice. SOUL.md is loaded on every turn. Core truths, boundaries, and vibe. Sharp beats vague, short beats long, corporate-mush kills agents.
MEMORY.md (long-term, injected every turn), daily notes (on-demand, free until searched), the dreaming pipeline that promotes durable facts, hybrid search.
The operating manual. Safety defaults, subagent spawning, the four-deep nesting model, and the three independent security mechanisms.
Tools are raw capabilities (exec, read, write, web). Skills are SKILL.md instruction manuals. Core 20-tool set, ClawHub marketplace, security audit pattern.
WebHooks, cron, and HEARTBEAT together. GoGo skill for Google Workspace, SearXNG vs Brave for search, Playwright browser automation with SSRF blocking.
Containerize. Sandbox modes (off, non_main, all), volume mounts, gVisor for kernel isolation, Docker bypassing UFW, the production hardening checklist.
Hetzner, DigitalOcean, or Linode. The curl installer with --install-daemon, Tailscale for remote access, openclaw security audit --deep, log rotation.
Stop using your agent as a chatbot. Align-before-build prompting, build-log files for multi-session continuity, the orchestrator pattern with research subagents.
One gateway, many agents. Per-agent workspaces, agent directories that must never collide, deterministic routing precedence, agent-to-agent communication.
Every workspace file, every openclaw.json section, the full bootstrap injection order, and the end-to-end message routing pipeline.
Read or watch? Use both.
The video is the recommended primary medium because the demos (terminal commands, dashboard tours, channel pairing) make more sense in motion. The written pages at /bootcamp/day-N are the recommended reference afterward, because the exact config snippets and copy-pasteable commands live there.
The pattern most builders settle into: watch the chapter, switch to the written page for the homework, jump back into the video for the next chapter.
OpenClaw Video Course FAQ
Where do I watch the OpenClaw video course?
Right here on this page, embedded above, or directly on YouTube at https://youtu.be/lyZSFSe3Gj0. The video is one continuous 4-hour recording covering all 16 lessons back-to-back, with hard chapter markers at the start of each day so you can jump directly to the lesson you want.
Is the OpenClaw video course on YouTube?
Yes. The full 4 hours 20 minutes course is published on YouTube and embedded on this page for in-place watching. No paywall, no signup, no email gate. Standard YouTube playback controls work, including 2x speed, captions, and chapter jumps.
Can I watch the OpenClaw video course at 2x speed?
Yes, that is the recommended pace for first-time watchers who already know what AI agents are. 2x cuts the runtime in half. Slow back down to 1x for Day 4 (cost optimization), Day 9 (AGENTS, security mechanisms), and Day 13 (VPS deployment) where the configuration syntax matters and small details bite.
What if I don't have 4 hours? What should I watch first?
If you have 30 minutes, watch Day 1 and Day 2 for the mental model and the install. If you have an hour, add Day 5 (channels) and Day 7 (SOUL.md) for the personality layer. If you have 90 minutes, add Day 6 (HEARTBEAT.md) for proactive scheduling. By that point you have an agent that talks back on Telegram with your voice and runs morning briefings on its own.
Are there chapters in the OpenClaw video?
Yes. Each of the 16 days starts at a hard chapter marker inside the video. The chapter list is in the YouTube description, and this page also has a day-by-day index where each row deep-links to the exact second of the YouTube video where that day starts.
Can I download the OpenClaw video course?
The video lives on YouTube under standard YouTube terms. For offline watching, YouTube Premium offers offline downloads on mobile, or pin the video for later. The accompanying written lessons at openclawconsult.com/bootcamp work fine offline once cached.
Are there subtitles or transcripts?
YouTube auto-generates English captions for the video. For a cleaner reading experience, every video lesson is also published as a long-form written page at /bootcamp/day-N with the same material in text form. Many viewers prefer reading the day page first and then watching the video for the live demos.
Why a single 4-hour video instead of 16 separate uploads?
Because the lessons compound. The single-video format makes it easy to jump from "I just finished Day 5" to "now show me Day 6" without leaving the player or breaking the build-along context. The chapter markers give you the same jump-anywhere experience as a playlist would, with less friction.
Who recorded the OpenClaw video course?
Adhiraj Hangal, founder of OpenClaw Consult and author of openclaw/openclaw#76345, the cost-runaway circuit-breaker fix merged into OpenClaw core by project creator Peter Steinberger in May 2026. The video is recorded in one continuous take per day and edited only for clarity.
After the video
Done watching? Open the written reference at /bootcamp and start working through the homework, one day at a time.
Need a deeper reference than the course covers? The OpenClaw Lab has 240+ articles on every aspect of the platform.
Want a production agent shipped under fixed scope in 2-4 weeks instead of building it yourself? Hire OpenClaw Consult.