Guide

Your AI Personal Assistant

People keep asking us how to set up an AI assistant. Here’s everything we’d tell you over coffee, minus the coffee.

A guide from the team at Boatwork · Updated April 2026

What Are We Actually Talking About?

An AI assistant is software that does real tasks — manages your calendar, drafts emails, researches, organizes files, and remembers context. It’s not a chatbot. It’s a worker.

📅

Calendar & Scheduling

Manages meetings, resolves conflicts, sends invites on your behalf.

Email & Drafts

Writes, edits, and sends emails matching your tone and style.

🔍

Research & Analysis

Pulls data, summarizes findings, and builds reports from the web.

🗒

Memory & Context

Remembers your preferences, past conversations, and ongoing projects.

The part nobody mentions upfront: setting up an agent is the easy part. The real work is keeping it running — models change, configs drift, skills break. If computers just worked, nobody would sell IT services. Same applies here.

Pick Your Path

Two fundamentally different approaches. One is managed — you use it, someone else runs it. The other is open — you own it, you run it.

Recommended

Claude Pro + Cowork

$20/mo

Powerful assistant, zero infrastructure. Anthropic handles everything.

💬
ChatResearch, writing, analysis with persistent memory
💻
CoworkWorks ON your computer — Gmail, Calendar, Slack, local files
📱
DispatchText a task from your phone, come back to finished work
🖱
Computer UseControls your screen when no direct connector exists
Setup15-30 min
MaintenanceNear zero
Cross-devicePhone, desktop, web
Trade-offVendor lock-in
Power Users

OpenClaw on a Mac Mini

~$600 to start

Own your stack. Any model, any channel, full data sovereignty.

🔄
Model FreedomClaude, ChatGPT, Gemini, Llama — swap anytime, fallback chains
🎭
Custom IdentityYour name, your voice, your brand — not a tech company’s
💬
Multi-ChannelWhatsApp, Telegram, iMessage, Discord, Slack — one agent
🔒
Full SovereigntyYour data never leaves your machine. 100+ extensible skills
Hardware~$499-599 (Mac Mini)
Ongoing API costs~$20-50/mo
Setup time2-4 hours
Maintenance1-3 hrs/month
Trade-offYou’re the IT dept

Three Paths, One Decision

1

Just get going

Claude Pro at $20/month. Desktop app. Phone paired. Running in 15 minutes.

Best if you want help now and don’t want to manage anything.
2

Own your stack

Mac Mini (~$500) + OpenClaw + API costs (~$20-50/mo). Your models, your persona, full sovereignty.

Best if you want independence and multi-model flexibility.
3

Have it managed

Open platform power without being your own IT department. Someone else handles it.

Best for businesses that need reliability every day. Learn more

Get Your Hardware

Going the OpenClaw route? You need a computer that stays on.

Apple Mac Mini M4

$499-599 retail
16GBUnified Memory
SilentFanless operation
HeadlessNo monitor needed
~5WIdle power draw

An M5 version is expected mid-2026. If you can wait, wait. Otherwise the M4 is excellent. After setup, tuck it on a shelf — you’ll interact via messaging apps on your phone or desktop.

Ongoing costs: ~$20-50/mo in API fees (depends on usage), ~$1-2/mo in electricity. No subscription for OpenClaw itself.

OpenClaw Quickstart

Seven steps from unboxing to a working agent.

1

Set up Mac Mini

Plug in, run macOS setup, connect to Ethernet.

2

Install OpenClaw

Open Terminal and run: curl -fsSL https://openclaw.ai/install.sh | sh

3

Run onboarding

openclaw onboard --install-daemon

4

Get an API key

Anthropic recommended. Or connect your existing ChatGPT/OpenAI account as an alternative.

5

Connect a channel

Telegram is easiest. The wizard walks you through it.

6

Test it

Send a message. If it responds, you’re live.

7

Go headless

Unplug the monitor. Your agent runs as a daemon 24/7. Tip: grab an HDMI emulator plug (~$8 on Amazon) so macOS thinks a display is connected — prevents resolution issues when you remote in.

First Week Playbook

Your agent is running. Here’s how to get real value out of it in the first seven days.

📝

Start with a system prompt

Tell your agent who it is and how to behave. Here’s a starter you can customize:

You are my personal assistant. Be concise, direct, and proactive. When I give you a task, confirm you understood it, then do it. If something is ambiguous, ask one clarifying question — don’t guess.
📧

Connect one tool at a time

Don’t wire up everything on day one. Start with calendar or email, get comfortable, then add Slack, file access, and others.

💪

Give it a real task daily

“Summarize my unread emails” or “Draft a reply to Sarah’s proposal.” The only way to learn what works is to actually use it.

🔎

Review the logs

Check what your agent did and didn’t do. Tune the instructions based on what surprises you. This is where the real value gets built.

Test safely first

Before granting write access to email or files, test with read-only permissions. Make sure the agent does what you expect before giving it the keys.

Security & Privacy Best Practices

Running your own agent means running your own security. Here are the essentials.

👤
Run as a limited userNever run your agent as root or admin. Create a dedicated macOS user account with only the permissions it needs.
🛡
Lock down the networkUse Tailscale or a strict firewall. Only expose the messaging channels you actually use — nothing else.
💾
Back up regularlyYour OpenClaw data folder (~/.openclaw) holds memory, config, and skills. Back it up weekly at minimum.
🔐
Limit tool permissionsBe extremely careful with tools that read/write email, access files, or execute commands. Start read-only, escalate slowly.
🛡
Watch for prompt injectionIf your agent processes external content (emails, web pages), malicious instructions can hijack it. Review agent logs periodically.
🔄
Keep everything updatedOpenClaw, Node.js, macOS — patch promptly. Open-source moves fast, and so do the vulnerabilities it fixes.

Book a One-Time Setup Session

We’ll get you set up over a video call. After that, you’re on your own. Pay securely via Stripe — you’ll get a confirmation email with scheduling details.

Guided Setup

$15030-minute video call

Screen share walkthrough for the tricky parts. You picked your path — we’ll get you unstuck.

Full Agent Build

$500Half-day video session

We handle it end-to-end: install, configure, connect accounts, build workflows, and deliver a maintenance cheat sheet.

Common Pitfalls

What we see people run into. Not fear-mongering — just experience.

🔄

“It worked last week”

Models update, APIs change, skills break. You need to stay current and test after every upgrade.

🧠

“My agent forgot everything”

Memory is fragile. A bad update or workspace reset wipes weeks of context. Backups are on you.

🤔

“Which model do I use?”

New models drop constantly. Picking the right one for the job is ongoing, not one-time.

🔒

“Someone got into my agent”

Prompt injection attacks are real. External content can hijack your agent’s tools.

👥

“Nobody uses it”

Business agents need workflow integration, data policies, and team-specific tuning.

“It takes more time than it saves”

If maintaining AI infrastructure becomes a second job, you’ve traded one problem for another.

If any of that sounds like your Tuesday, there’s a different option.

Boatwork Assist is a managed OpenClaw service for businesses. We run your AI agents so you don’t have to.

Upgrades handledOpenClaw updates, model migrations, skill patches
Integrations maintainedCRM, email, messaging — kept working when APIs change
Memory managedCorporate knowledge structured, agent tuned to improve
Data sovereigntyRight models, right tasks, sensitive data stays put
Custom identityYour agent represents your business, not a tech company
Always-on supportSecurity, monitoring, and incident response included

Marine industry? This is what we do at boatwork.co. Another industry? Reach out — we’re expanding.