Computer Club 2026

My AI Ghost
in the
Machine

How I built a personal AI assistant
that actually knows me

Rob Hofman · Microsoft Integration Architect

Context 01

ChatGPT
Forgets
Everything

Every session starts from zero. No memory of your projects,
your style, your preferences. Just a blank slate.

No memory No context No proactivity No personality
Platform 02

What is
OpenClaw?

  • Personal AI agent platform - self-hosted
  • Connects to Telegram, WhatsApp, Signal, Discord
  • Routes to multiple LLM providers
  • Runs tools: browser, shell, files, APIs
  • Extensible with custom skills

In one sentence

Your own AI assistant - running on your own machine, knowing who you are.

Setup 03

My
Setup

  • Ubuntu server on Tailscale
  • Telegram as primary channel
  • GitHub Copilot · synthetic.new · Claude Max
  • File-based memory system
  • Custom workspace with personality files
~/.openclaw/workspace/
├── SOUL.md        ← who I am
├── USER.md        ← who Rob is
├── MEMORY.md      ← long-term memory
├── HEARTBEAT.md   ← proactive checks
└── memory/
    └── 2026-03-22.md
Intelligence 04

The
Brain

Choosing the right model for the right task can significantly reduce costs.

Tier 1

GitHub Copilot

Tier 2

synthetic.new

Kimi K2.5 · GLM-4.7 · MiniMax M2.1

Tier 3

Claude Max

Identity 05

Giving it
a Soul

# SOUL.md
Be genuinely helpful, not performatively helpful.
Skip the "Great question!" - just help.

Have opinions. You're allowed to disagree.

Be resourceful before asking. Read the file.
Check the context. Search for it. Then ask.

Each session, you wake up fresh.
These files ARE your memory.

A markdown file that shapes personality, tone, and behaviour.

Continuity 06

Memory
that Persists

  • MEMORY.md - curated long-term memory
  • memory/YYYY-MM-DD.md - daily notes
  • Agent reads them every session before responding
  • Agent writes to them after important events
  • No database. Just files.

Key insight

"Mental notes don't survive session restarts. Files do."

Proactivity 07

Heartbeats
& Cron

  • Heartbeat polls every 2 hours
  • Checks tasks, emails, Obsidian notes
  • Messages Rob if something needs attention
  • Cron jobs for precise scheduling
08:30  Daily motivation
10:00  Morning check-in
13:00  Afternoon check-in
17:00  Wrap-up
16:00  Weekly summary
Real world 08
What we
built
Project 09

Cloud
Infrastructure

Designed and deployed a full Azure microservices platform - end to end.

40+ Azure resources 7 DevOps pipelines APIM · Service Bus · Logic Apps

How the agent helped

Ran pipelines, validated deployments, managed RBAC roles, tracked progress across sessions.

Project 10

Teaching
Materials

Generated a full course curriculum overnight - ready for class in the morning.

19 exercises Progressive difficulty Syntra curriculum

Time saved

What would have taken a weekend took one session while I slept.

Project 11

Work
Tracking

Automatic time tracking across all client projects - no manual entry.

Multiple clients Weekly summaries Obsidian sync

How it works

I mention what I worked on in conversation. The agent logs it, categorises it, and writes the weekly report.

Capabilities 12

What it
Can Do

  • Shell & file operations
  • Spawn sub-agents in parallel
  • Send & receive emails
  • Image generation
  • Manage tasks (Vikunja/todoist)
  • Read & write Obsidian notes
  • Custom skills
  • Multi-channel messaging
Lessons 13

What
I Learned

  • Choose the right model for the right task - it saves on costs
  • Personality files matter more than the model you use
  • Memory = files. Simple beats clever every time.
  • The hard part isn't the AI - it's knowing what to ask for
  • Having it reach out to you > you reaching out to it

Core insight

SOUL.md turns a generic LLM into your assistant.

Get started 14

Try it
yourself

npm install -g openclaw
openclaw init
openclaw gateway start
docs.openclaw.ai github.com/openclaw/openclaw clawhub.com

Community

discord.com/invite/clawd

Thanks 15
Q&A

Rob Hofman · Computer Club 2026