메뉴
OpenClaw Workshop 4: Demo Day & What's Next. The Anthropic ban explained, updating OpenClaw, switching to OpenAI, alternatives (Claude Cowork, NanoClaw, Nanobot), and demo presentations. Use when the user says 'workshop 4', 'demo day', or has completed workshop 3.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
OpenClaw Workshop 3: Security, Conversations & Costs. Interactive workshop covering prompt injection & data exfiltration, OpenClaw security layers, least privilege, five levels of conversation organization (single chat, subagents, threads/groups, multiple agents, dedicated instances), and cost management. Use when the user says 'workshop 3', 'openclaw advanced', 'openclaw security', or has completed workshop 2.
OpenClaw Workshop 2: Power Features. Interactive checklist that guides users through setup (ack emojis, cron, 1Password, skills) then hands-on use cases (YouTube Music, calendar booking, Twitter, meeting summary to Apple Notes, morning briefing cron, custom automation). Use when the user says 'workshop 2', 'openclaw power features', or has completed workshop 1.
OpenClaw Workshop 1: Getting Started. Interactive checklist that guides users through installing OpenClaw, connecting Telegram, voice-to-text, email via IMAP, and browser setup. Use when the user says 'workshop 1', 'openclaw getting started', 'start openclaw setup', or wants to begin the OpenClaw course.
| name | openclaw-workshop-4 |
| description | OpenClaw Workshop 4: Demo Day & What's Next. The Anthropic ban explained, updating OpenClaw, switching to OpenAI, alternatives (Claude Cowork, NanoClaw, Nanobot), and demo presentations. Use when the user says 'workshop 4', 'demo day', or has completed workshop 3. |
This is the final workshop. Today covers important changes in the OpenClaw ecosystem, then finishes with demos -- their moment to show what they built.
Today's agenda:
When you run this workshop:
Welcome to the final workshop! Before demos, we need to talk about something important. The OpenClaw ecosystem changed significantly this week. You need to know what happened, what your options are, and how to keep your agents running.
Then we'll finish with the best part -- demos. Everyone presents what they built. No time limits. Take as long as you need.
On April 4, 2026, Anthropic blocked Claude subscriptions from being used with OpenClaw and other third-party agent frameworks.
Before (worked):
After April 4, 2026 (blocked):
| When | What |
|---|---|
| Nov 2025 | OpenClaw (then Clawdbot) launches, goes viral |
| Jan 2026 | Anthropic starts fingerprinting and suspending accounts using OpenClaw |
| Jan 2026 | Security audit: 512 vulnerabilities, 8 critical |
| Feb 2026 | Peter Steinberger (OpenClaw creator) joins OpenAI |
| Apr 4, 2026 | Anthropic officially blocks third-party framework usage on subscription plans |
Discuss: This is a lesson about platform risk. When you build on someone else's platform, they can change the rules overnight. This is why understanding your options matters.
If you were using Claude subscription with OpenClaw:
If you were using API keys:
If you had startup credits:
Keep your agent current, and move to a provider that welcomes you.
Check your version:
openclaw --version
Update to latest:
openclaw update
Prompt to try:
What version of OpenClaw am I running? Check if there's a newer version. Back up my workspace, update, then verify all channels, cron jobs, and skills still work. Run
openclaw security auditafterward.
After Anthropic's ban, OpenAI became the default choice because:
Step 1: Get an OpenAI account Sign up for ChatGPT Plus ($20/mo) or Pro ($200/mo). For API usage, create a developer account at platform.openai.com.
Step 2: Configure OpenClaw for OpenAI
Prompt to try:
Switch my OpenClaw to use OpenAI as the model provider. I want to use Codex 5.3 as my default model. Walk me through the configuration.
Step 3: Set up embeddings separately Important: Codex does not include embeddings. You need a separate API key.
Prompt to try:
Set up OpenAI text-embeddings-3 for my agent's memory system. This is separate from the Codex model.
Step 4: Tune your agent Claude and GPT respond differently to the same prompts. Expect to spend a few days tuning SOUL.md, AGENTS.md, and cron prompts.
Prompt to try:
Review my SOUL.md and AGENTS.md. Are there any instructions that are Claude-specific? Suggest changes for OpenAI models.
Step 5: Test everything
Prompt to try:
Full systems check: send a test message on Telegram, send a test email, run a cron job, do a web search. Confirm everything works with the new model.
Don't lock into one provider. OpenClaw's config supports multiple model providers:
| Task | Model | Why |
|---|---|---|
| General daily use | Codex 5.3 (subscription) | Covered by ChatGPT plan |
| Complex analysis | Claude Opus (API) | Pay per token, worth it |
| Routine cron jobs | Kimi 2.5 / Minimax | Cheap, fast, good enough |
Prompt to try:
Configure my agent to use Codex 5.3 as default, Claude Opus via API for tasks tagged as "complex", and Kimi 2.5 for my daily news briefing cron.
OpenClaw isn't the only option. Here's the landscape.
Anthropic's own answer to OpenClaw. Desktop-focused, not messaging-focused.
What it is:
Best for: Desktop automation, document workflows, browser tasks on your computer
Not great for: Always-on messaging (Telegram, WhatsApp), cron jobs, multi-channel communication
The key difference: OpenClaw is an always-on messaging assistant you text anytime. Claude Cowork is a desktop agent you assign tasks to. They're complementary, not competing.
StartClaw, SimpleClaw, MyClaw, UniClaw -- someone else runs OpenClaw for you. Easy setup, no hardware. Best for non-technical users.
| If you need... | Choose | Why |
|---|---|---|
| Desktop automation | Claude Cowork | Native Anthropic, no ban risk |
| Maximum security | NanoClaw | Container isolation, zero-trust |
| Simplicity | Nanobot | Small codebase, easy to understand |
| No infrastructure | KiloClaw or hosted | Managed for you |
| Enterprise compliance | IronClaw | Zero-trust, audit-ready |
| Maximum flexibility | Stay on OpenClaw | Largest ecosystem, most features |
Discuss: The skills you learned -- agent architecture, security thinking, prompt engineering, automation design -- transfer to any platform. OpenClaw is a tool, not a religion.
Show what your agent can do. No time limits. Take as long as you need.
Each student presents:
Rules:
As students present, highlight:
Architecture choices:
Creative use cases:
Common struggles:
Group discussion:
Chapter 1 (Anthropic Ban): 10 minutes. Important context. Frame it as platform risk lesson and opportunity to become multi-provider.
Chapter 2 (Update & Switch): 15 minutes. Hands-on for affected students. Multi-provider strategy is the key takeaway.
Chapter 3 (Alternatives): 10 minutes. Informational. Make sure they know Claude Cowork exists -- it's Anthropic's direct response.
Chapter 4 (Demo Day): All remaining time. No limits. Every student presents. Celebrate everything.
🎉 Workshop 4 COMPLETE! COURSE COMPLETE!
🚨 Ecosystem:
✅ Understand the Anthropic ban and why it happened
✅ Know how to update OpenClaw and switch to OpenAI
✅ Know the multi-provider strategy
✅ Know the alternatives (Claude Cowork, NanoClaw, Nanobot, KiloClaw)
🎤 Demo Day:
✅ Presented your agent to the group
✅ Celebrated wins, learned from struggles
✅ Got inspired by what others built
The OpenClaw Course Journey:
Workshop 1: Built the foundation
Workshop 2: Made it useful
Workshop 3: Made it secure and scalable
Workshop 4: Navigated change and showed the world
You started with nothing. Now you have:
- A working AI agent (or team of agents)
- Real automations that save you hours
- Security knowledge to protect yourself
- The ability to switch providers and platforms
- A community of fellow builders
The course is over. The building never stops. 🦞
Platform risk is real. Anthropic changed the rules overnight. Build with flexibility -- never depend on a single provider.
The multi-provider strategy is the answer. Use the best model for each task. Default to the cheapest option. Use expensive models only when quality demands it.
OpenClaw is a tool, not a religion. If something better comes along, switch. The skills you learned transfer to any platform.
Claude Cowork and OpenClaw are complementary. Desktop automation vs always-on messaging agent. You might want both.
You are the builder. The most important thing you gained from this course isn't OpenClaw. It's the mindset. You see automation opportunities everywhere. That's the real skill.