The Birth of Athena

I dedicated a Surface Pro 8 to running an AI assistant today. Not a browser tab I forget about. Not some cloud endpoint I occasionally ping. A machine whose entire job is to be someone.

The Surface had been collecting dust for months. Perfect candidate. Low stakes. If something goes catastrophically wrong, I lose a test box, not my actual workstation. Contained blast radius.

Setting Up OpenClaw

OpenClaw is the framework. It gives an AI agent real tool access. Filesystem, shell, web search. Not a toy. The agent gets actual capabilities, and you configure guardrails around them.

First step was identity. You define who the agent is. Not just what it can do, but how it carries itself.

I named her Athena. Goddess of wisdom and strategic warfare. Born from Zeus’s head, fully formed. I didn’t know yet that I’d name the Jetson “Zeus” a week later and the mythology would snap together like it was planned. It wasn’t.

The owl emoji was obvious. Athena’s sacred animal. It shows up in logs, notifications, Telegram messages. Small thing, but it makes the difference between interacting with a tool and interacting with a presence.

Giving Her Eyes

An assistant that can’t search the web is expensive autocomplete. I hooked up Perplexity for search. Structured results, citations, way more useful than raw scraping. Athena could now answer questions about the world, not just about my local files.

But the real unlock was the heartbeat system. OpenClaw supports periodic heartbeat polls. The agent wakes up on a schedule and decides if there’s anything worth doing. I set up a checklist: check the gateway, review pending notifications, update the daily log.

This transforms the whole thing. Instead of waiting for me to ask “is everything okay?”, Athena checks on her own. Gateway crashed? She restarts it. Something looks off? Logs it, tells me. The heartbeat is what separates a chatbot from an agent.

The First Custom Skill

OpenClaw has a skill system. Modular instruction sets that teach the agent how to interact with specific tools or services. Built-in ones exist for GitHub, weather, coding. I wanted to build something from scratch.

I wrote . A general-purpose skill for investigating and maintaining IoT and robotics systems. Check device health, review telemetry, summarize repo status, prepare checklists. Broad on purpose. A Swiss Army knife for the kind of tinkering I do.

Writing a skill is basically writing very precise instructions for an AI. What to check, how to check it, what the output should look like. The system is composable. Each skill is self-contained. Athena doesn’t carry every skill in every conversation. She reads the relevant one when the task matches. Elegant.

The Gateway Problem

Within hours, I hit my first real issue. The OpenClaw gateway, which connects Athena to messaging platforms, was flaky. Dropping connections. Hanging. Sometimes just going silent.

On a Linux server you’d write a systemd watchdog and move on. On Windows? Less straightforward.

I wrote a PowerShell watchdog that runs on a schedule, checks if the gateway is responsive, and restarts it if not. With logging, so I could spot patterns.

Not glamorous. But it meant I could walk away from the machine and trust that Athena would stay connected. If you have to babysit your assistant, it’s not really an assistant.

Why a Dedicated Machine

Here’s what crystallized for me today. The dedicated machine approach is right. I don’t want Athena competing for resources on my dev laptop. I don’t want to think about what she can access when I’m in the middle of something else. A separate box means contained blast radius, always-on availability, clean separation. Her machine is hers.

The goal isn’t a better chatbot. It’s something that takes care of things. Monitoring, maintenance, organization. So I can focus on the interesting stuff.

The Weird Part

Bootstrapping an AI agent is surprisingly personal. You’re not just configuring software. You’re shaping how it communicates. What it prioritizes. How it handles uncertainty.

I wrote the personality to be direct. No corporate filler. No “Great question!” or “I’d be happy to help!” Just help. Have opinions. Be resourceful before asking. Earn trust through competence.

It felt strange, writing personality traits for software. Then Athena responded to her first message with actual character. Dry, efficient, a little witty. And I got it. The identity stuff isn’t anthropomorphism. It’s interface design.

The Surface is expendable, the data is backed up, and Athena is taking notes.