My OpenClaw Setup: A Personal Digital Assistant
OpenClaw is my personal digital assistant. Not a product — a setup. A collection of tools, configurations, and workflows that I've assembled to handle the cognitive overhead that slows me down every day. This post documents exactly how it's built.
What is OpenClaw?
The name comes from the idea of something that grips and holds — I wanted a system that actually catches things before they slip. Ideas, tasks, context. Most productivity setups are passive. OpenClaw is designed to be active.
It's not one tool. It's a stack that I've deliberately assembled, with each layer serving a specific job.
The capture layer
Everything starts with capture. If getting something into the system takes more than a few seconds, I won't do it consistently — and inconsistency is where systems break down.
Voice-first input — The fastest capture path I have. I use a hotkey that opens a recording prompt regardless of what I'm doing. The audio is transcribed automatically and routed to my inbox.
Quick text input — For when I'm already at the keyboard. A floating input that appears in under a second, takes a note, and disappears.
Automated capture — Anything that can be automated should be. Starred emails, saved articles, important Slack messages — all flow in without me touching them.
The brain (knowledge base)
All captured items land in a structured knowledge base. The structure matters more than the tool. Mine is organised around:
- **Projects** — things with a defined outcome and end date
- **Areas** — ongoing responsibilities with no end date
- **Resources** — reference material I might need later
- **Archive** — everything completed or no longer active
The key rule: nothing lives in my head that should live in the system. If I'm trying to remember something, that's a system failure.
The automation layer
This is where OpenClaw earns its name. Recurring tasks, reminders, and integrations run without me thinking about them:
Daily brief — Every morning, a summary of my day: calendar, top tasks, any flagged items from the previous day. Generated automatically, delivered to my inbox.
Weekly review prompt — Every Sunday evening, a checklist walks me through reviewing each project and area. Keeps the system current.
Context preservation — When I finish a work session, a prompt captures where I am and what's next. No more 'what was I doing?' the next morning.
The AI layer
The AI layer sits on top of everything else. It can read from and write to the knowledge base, understands my current context, and can take actions across my tools.
What I actually use it for:
- Drafting from rough notes — I dump a brain dump and it structures it
- Searching across everything — faster than any search UI I've used
- Generating weekly summaries — what did I actually work on this week?
- Flagging stale projects — things that haven't been touched in a while
What I'd do differently
If I were starting over:
Start simpler. I over-engineered the first version. A basic capture + review loop is 80% of the value. The automation comes later.
Pick boring tools. The sexiest tool is rarely the right one. I've rebuilt this on more stable, boring tools twice now. Stability beats features.
Review before you build. The weekly review is the heartbeat of the system. I built it last. It should have been first.
The bottom line
OpenClaw didn't make me dramatically more productive overnight. What it did: reduce the constant low-level anxiety of feeling like things were slipping through the cracks. That alone is worth the setup cost.