OpenClaw.
The course that puts AI in your pocket.

Explain what OpenClaw is and how it differs from ChatGPT, Claude.ai, and Cowork using the “three rooms of AI” framework (browser room, desktop room, pocket room).

Key takeawayOpenClaw isn’t a chatbot — it’s an AI that lives on your devices and speaks through your apps.

Evaluate whether local-first AI fits your privacy and workflow needs using a 5-question decision framework, and understand why Anthropic cut subscription access on April 4, 2026.

Key takeawayLocal-first trades a little convenience for a lot of control.

Analyze which tool to reach for in five common work scenarios using the tool-choice decision matrix.

Key takeawayDifferent AIs for different rooms — OpenClaw is the one you keep in your pocket.

Evaluate four hosting options (laptop / Mac Mini / Raspberry Pi / VPS) against four criteria (always-on, privacy, cost, local-model capable) and choose the right one.

Key takeawayIf it needs to run while you sleep, it can’t live on a laptop that sleeps.

Evaluate the three model options (cloud API / competitor API / local Ollama) and choose the right primary model for a working professional’s use case, given the April 4, 2026 subscription cutoff.

Key takeawayThe best model is rarely the biggest — it’s the one matched to your job, your wallet, and your privacy.

Install OpenClaw on your chosen hardware using Docker and verify the gateway is running using openclaw doctor --deep --yes.

Key takeawayInstalling OpenClaw is closer to installing Slack than to compiling code — if doctor is green, you’re in.

Send your first three requests to OpenClaw via Telegram and evaluate the quality of each against a simple rubric (correct, useful, context-aware).

Key takeawayYour first OpenClaw session should ship a real output, not introduce itself.

Write requests in the messaging-short style that gets better OpenClaw responses than full-paragraph prompts.

Key takeawayOpenClaw lives in your messaging app — write for it like you’d write to a human there.

Design and ship a simple weekend-briefing automation (triggers Sunday 6 PM, summarizes next week from calendar, sends to Telegram).

Key takeawayA working L1 automation is the moment OpenClaw becomes yours — not a toy, a tool.

Configure OpenClaw’s memory system AND the new memory-wiki (April 2026) to persist facts and documents across sessions using the three memory types (user, project, feedback).

Key takeawayA memory-less AI is a stranger every morning. Give yours a notebook AND a filing cabinet.

Use OpenClaw’s session branching feature to fork a conversation mid-flow, try an alternate path, and merge or abandon.

Key takeawayThe best prompt is often the second attempt — branching lets you have both.

Install a pre-built skill from ClawHub AND author a simple SKILL.md file that teaches OpenClaw to draft Sofia’s Monday standup in her team’s exact format.

Key takeawayA skill is just a markdown file with instructions — if you can write an email, you can write a skill.

Configure a cron job that runs the weekly-recap skill every Monday at 8:30 AM and sends the output to Telegram.

Key takeawayCron turns OpenClaw from reactive to proactive.

Build a TaskFlow that triggers when a specific Gmail arrives (e.g., from your manager) and executes a multi-step chain: summarize, draft reply, notify in Telegram.

Key takeawayTaskFlows connect OpenClaw to the moments that actually matter in your day.

Use OpenClaw’s live Canvas to review agent actions in real time, and enable on-device voice mode (April 2026 MLX feature on Apple Silicon) for hands-free sessions.

Key takeawayNever let an agent act without a window you can watch it through.

Connect OpenClaw to your primary messaging app(s), set up channel routing so work and personal agents stay separated, AND add an Apple Shortcuts trigger for one-tap home-screen actions.

Key takeawayOne agent, many channels — but give each channel a clear job.

Configure OpenClaw’s MCP connectors for Google Calendar, Gmail, and Slack to feed real context into your automations.

Key takeawayAn AI without your context is a generic chatbot. Give yours the keys to your stack.

Configure Tailscale to expose your OpenClaw gateway to your own devices (and ONLY your own devices) without opening any ports to the public internet.

Key takeawayYour agent should only be reachable by you — Tailnet makes that effortless.

Analyze three real ClawHub skills to identify the four reusable design patterns (linear, branching, verification, delegation) and pick the right one for a new task.

Key takeawayThe best skills aren’t the longest — they’re the ones that match the shape of the job.

Build a blocker-detector skill that reads Slack, identifies real blockers, filters noise, and pings only on true positives — handling three common failure modes.

Key takeawayProduction skills get the filter right — the goal isn’t to ping more, it’s to ping better.

Orchestrate a main agent plus research sub-agent pair using OpenClaw’s updated sub-agent config — nested agents no longer block each other.

Key takeawaySub-agents let you decompose problems the way real teams do.

Layer Paperclip on top of multiple OpenClaw agents to create an org chart, assign budgets, set goals, and track cross-agent work in one dashboard.

Key takeawayOpenClaw gives you employees. Paperclip gives you a company.

Configure OpenClaw’s router to send simple tasks to cheap models, complex tasks to flagship models, and private tasks to a local model — optimizing for cost, speed, and privacy.

Key takeawayThe best model is rarely the biggest — it’s the one matched to the job’s shape.

Install Ollama plus a local model and wire it to OpenClaw for tasks that must never leave your device.

Key takeawayLocal models are your “never leaves the house” option — not your default.

Evaluate OpenClaw’s six most common failure modes (hallucination, stale memory, wrong-tool, silent cron failure, leaked context, prompt injection) using real incident logs and design a guardrail for each.

Key takeawayThe question isn’t whether your agent will fail — it’s whether you’ll notice when it does.

Design a personal-use governance policy for OpenClaw covering data residency, key management, audit logs, rotation schedule, and a tested “kill switch” procedure.

Key takeawayThe day you need a kill switch is too late to design one — write it while nothing’s wrong.