AI / Artificial Intelligence Automation & Workflows

OpenClaw 🦞 - Autonomous AI Agent in 1 Day

From installation to a working agent in 7 hours. Build your AI assistant that works 24/7 and does what you never have time for (but always dreamed of ;)

OpenClaw Agent handling tasks for Lucas Matuszewski in modern cyberpunk office

Goals

Install and configure OpenClaw on your own device
Create a working agent tailored to your needs
Master: Heartbeats, Cron Jobs, Skills, Memory, Web Search
Understand costs, risks, and best practices

Course content

Below is a sample training program for 1-2 days. Session durations shown in titles are approximate. We always customize the program to the group's needs and business objectives.

History: Clawdbot → MoltBot → OpenClaw + OpenAI foundation
180k stars, Google trends, Peter + OpenAI
How it differs from Claude Code, Codex, Copilot
"Agent as Code" - what autonomy really means
When to use, when NOT to use
Hands-on: Start installation (downloads in background)
Architecture: gateway, agent, workspace, session, files
Installation via npm + TUI wizard (Docker and VPS are optional advanced modules)
Remote Dashboard access: Tailscale + SSH (sshfs option)
Security: separate user, minimal permissions, backups, etc.
Verification: everyone has OpenClaw running
MD files: IDENTITY, SOUL, USER, AGENTS, MEMORY, HEARTBEAT
LLM connection (providers from survey): Opus 4.6, GPT-5.3, GLM-4.7
Messenger (from survey): WhatsApp/Slack/Teams/...
Test: you send a task → agent responds → success 🎉
Heartbeats vs Cron: orchestrator vs worker
Memory system: daily logs, knowledge files, RAG, search
Web Search: Perplexity/Brave/Grok (from survey)
Skills (1 example from survey): X.com/MoltBook/blog/...
MCP (1 example from survey): GitHub/Linear/Gmail/...
Hands-on: Agent has Heartbeats, memory, 1 Skill, 1 MCP
Hands-on: complex task from your use case
Dashboard: logs, debugging, common issues
Costs: tokens, cache, optimization (different models), context window
Security: summary of rules, ClawHub Skills (prompt injection!), sandbox, GDPR
What's next: materials, roadmap, community, advanced training
Q&A
>> OPTIONAL MODULES ON SPECIAL REQUEST:
Only if survey shows demand or on special request:
Why VPS/Docker: always online, better resources, isolation
VPS setup (Hetzner): specs, Ubuntu, SSH, Tailscale, firewall
Docker setup: Dockerfile, Compose, volumes, networking
File access: SSHFS/rclone, nvim, Fresh IDE TUI
2-3 additional Skills or MCPS based on the pre-course survey
Custom Skills: structure, we build one from scratch (instructions, scripts, assets)
Sub-agents: delegation, different LLM models for different roles, separate session (context)
Choice based on the pre-course survey - one in depth, examples:
Developer Assistant: code review, auto PR, refactoring, documentation, CI/CD
Project Manager: GOALS.md + PLANS/, Linear/Jira, tracking, reporting
Content Creator: research, drafting, SEO/GEO, publishing, analytics
Multi-agent mode - dedicated agent for the task or personalized agent for each team member:
Configuration and dedicated settings for each agent
Sendbox, data & memory separation, per-agent workspace
Communication rules for many agents (e.g. WhatsApp phone number routing)
Threat model: prompt injection (ClawHub case study), tool abuse, hallucinations
Defense: sandboxing, allowlist, filtering, monitoring, audit logs
GDPR: personal data, EU region, DPA, retention, provider ToS
Best practices for companies: separate environments, no prod access, tests as gates
Dashboard: usage metrics, token consumption, cache hit rate, error rate
Cost optimization: context engineering, multi-model, batch ops
Example: agent optimized itself (48% reduction)
Evals: success rate, time, cost per task, A/B testing
Observability: structured logging, tracing, debugging, post-mortem

Requirements

1. Dedicated device for the agent

Old laptop (Linux/macOS, optionally Windows)
Mac Mini
Optional: VPS or Docker (optional module, depending on needs)

2. LLM token source

Provider API: Claude Opus 4.6, GPT-5.3 Codex, GLM-4.7 or 5 (cheaper alternative from China)
Subscriptions (cheaper, but risk of account suspension): Claude Code, ChatGPT Plus + Codex CLI
Local models (Ollama/LMStudio) - if you have a powerful GPU and a good model with function calling

3. Search engine API (optional, but very useful):

Perplexity API (recommended)
Brave Search (free tier)
Grok Search

4. Tailscale account (free) - for secure remote connection to your agent

5. Messenger - choose one, e.g.:

WhatsApp (recommended, own number or dedicated)
Slack, Teams, Telegram, etc.

One week before the workshop you'll receive:

Installation and requirements guide
Survey: Your experience, expectations and preferences
We personalize the workshop to the group's needs

Target Audience

For those who want to see the future of AI-powered work today:

Developers - ways to automate your work
Tech leads and CTOs - potential and risks of autonomous agents in teams
Project managers and CEOs - assistant for task management and communication
Marketing and analytics professionals - automated research, writing, and publishing
AI enthusiasts who've heard about OpenClaw and want hands-on experience

IMPORTANT: The workshop assumes basic familiarity with the terminal (bash/zsh) and a solid understanding of computers and networking basics. We'll explain everything step by step, but for non-technical people the training may be challenging (but that's what we have AI for ;)

OpenClaw is for you if:

You waste hours on repetitive tasks (code review, documentation, task management)
You want to see how an autonomous agent works in practice
You're not afraid to experiment and learn new things
You understand it's early stage, doesn't work perfectly, and requires some tinkering

Not for you if:

You're looking for a "plug and play enterprise solution" (not yet)
You don't want to deal with configuration and troubleshooting (learning curve, time investment)
You're afraid to give AI access to tools and your system

Training Methodology

Hands-on workshop. Everyone leaves with a working agent.

Step-by-step installation (or come with it already set up)
Configuration from scratch
First tasks in real-time
Debugging and optimization
Materials and examples to take home

Additional information

Why OpenClaw? Why now? 🦞

December 2025. Austrian developer Peter Steinberger publishes "Clawdbot" - an autonomous AI agent that can do almost anything:
- Read emails and send Slack notifications
- Write code
- Manage projects
- Publish blog posts
- Even compile programs and add them to PATH
January 2026. The project explodes as "MoltBot" (Anthropic decided "Clawd" sounded too much like "Claude")
February 2026. Now as "OpenClaw", it becomes the fastest-growing project in GitHub history (180k stars in 2 months)
February 16, 2026. OpenAI announces that Peter Steinberger is joining the Codex team and will lead autonomous agent development. OpenClaw becomes an independent foundation, supported by OpenAI.

What does this mean?

Autonomous agents are no longer science fiction. These are systems that:

Plan work in hours, not minutes
Delegate tasks to other agents (sub-agents)
Learn, test, improve their own code and prompts (instructions)
Communicate via WhatsApp, Slack, Teams, etc.
Work 24/7, no vacation, no complaints

Hard numbers from 10 days working with OpenClaw (our case study):

893M tokens, 227 sessions, 11,297 messages, 8,499 tool calls
5 custom Skills created by the agent for itself
9 applications (CLI in Rust, Tauri notepad, monitoring)
10 Pull Requests to Edukey's production repo
Cost: ~$40 (using Claude Pro, Google AI Pro, ChatGPT Plus subscriptions)

The truth that hurts: When agents write code faster than you can do code review, you become the bottleneck. Your role changes. From executor to manager. From coder to strategist.

At this workshop:

We don't pretend it's "enterprise-ready" (it's not, yet)
We don't hide the risks (they exist: permissions, prompt injection, GDPR, costs, etc.)
We don't show slides (90% hands-on)

Instead, we'll show you:

How to install and configure OpenClaw securely
How to build an agent that actually works for you
How to control costs, debug problems, optimize workflows
How to think about autonomous agents in the context of your work and team

Every training is customized to participant expectations.

What's included:

Original training materials (PDF + repo with examples)
Hands-on labs - ready-to-use configs and scripts
Pre-training survey - we personalize the program for you
Certificate of completion
Ability to ask 2-3 questions up to 6 months after training (via email)
Evaluation: post-training survey with report

I you want to learn more about potential the Autonomous Agents have, read how we tested OpenClaw for 10 days with Opus 4.6 model from Anthropic in programming and other automations.