OpenClaw+GLM-5: Predictable Costs, Somewhat Secure (But Not Worth the Setup)

Most AI assistants either burn through your wallet (some spent hundreds a day) or feel like diet versions of the real thing (Gemini, powerful yet restrictive). OpenClaw with GLM-5 changed that math for me.

I've been running it for the past few weeks, and here's what actually matters: predictable costs, extensible skills, and a setup I can audit without wondering what's happening behind the scenes, especially with ocean of security concerns.

In the article, I’ll go over my use case of using OpenClaw as my personal assistant, high-level setup, the great and the ugly sides of OpenClaw.

I use Hostinger VPS to host OpenClaw. You can use an old laptop with no access to your personal accounts, but I had some Hostinger credits from my abandoned n8n projects, so I reformatted it with OpenClaw image. Easy setup process.

$5 per month (KVM1 is more than enough for normal usage). Get another 20% of discount via my referral link.

I'm using zai/glm-5 as my primary model in OpenClaw.

GLM-5 is an open source model that punches way above its weight. Fast enough for real-time responses, smart enough to handle context switching across tasks, and cheap enough that I don't think twice about using it. Best of all, it has a predictable subscription pricing instead of using Claude/OpenAI’s API usage-based pricing.

Cost breakdown: around $9/month for Lite, but I opted in for Pro, anticipating slightly heavy usage. I have it summarize my newsletters and do research for me, and it takes up about 80% of my quota during my first week.

Get another 10% off with my referral link.

I tried Whatsapp and Telegram, but ended up using Discord instead. It helps that I already have Discord desktop app, and I don’t want Bob (my AI Assistant) to have my phone number 😇.

As you’re onboarding, I also asked Bob to make security his core memory and principle. As you can tell, I’m seriously paranoid!

OpenClaw's skill system is what makes this whole thing work. I'm pulling from VoltAgent/awesome-openclaw-skills, which has 2,500+ cleaned-up community skills across every category you'd need.

The catalog is massive:

I don’t even know what most of them do. I simply ask Bob to cherry-pick what matters for my workflow as an executive assistant setup.

High Priority Skills Bob Picked for me:

Before adding any Skill, I vet it. I ask Bob to read the SKILL.md file, check what it actually does, and make sure I'm not importing something sketchy. Be careful of using any random Skill because some could be malicious and designed to retrieve your personal information.

I subscribe to 30+ newsletters. Tech, AI, startups, product management. They pile up every morning like homework I actually want to do but never have time for.

Here's the problem: I can't just ignore them. These newsletters contain signals for me to stay on top of trends. Product launches. Industry shifts. AI destroys jobs. But reading them all takes 20+ minutes I don't have.

The Workflow I Built:

Step 1: Set up Gmail OAuth access (this was the annoying part, but only had to do it once)

Step 2: Configure the newsletter digest workflow

Step 3: Run it every morning at 7am

It’s pretty magical. What used to take 45 minutes of skimming now takes 5 minutes of review. I'm not missing important signals, and I'm not drowning in redundant coverage of the same news.

The best part about this is Bob knows me even more now, because it knows what I read.

Bob has more personality than I expected. In fact, I could have given it more personality if I wanted to during onboarding (or edit SOUL.md file). When I uploaded my resume so it knows me better, Bob's response was:

😂 It's not trying to be a corporate chatbot with canned enthusiasm. It reacts like a human collaborator who actually looked at what you sent.

This kind of interaction matters more than you'd think. When you're triaging 100+ emails at 11pm, having an assistant that feels like a person instead of a FAQ bot keeps you engaged.

Three reasons:

I'm not saying OpenClaw is perfect. The skill ecosystem is still maturing, and you need to be comfortable setting up OAuth apps, hooking up API keys, and possibly spend money on peripheral skills. But if you're a builder who wants an AI executive assistant without the vendor lock-in and extreme customization, this setup is worth trying.

GLM-5 is fast, capable, and open source. OpenClaw is extensible and transparent. Together, they're a stack that is powerful and cost efficient.

Real talk: OpenClaw isn't worth the setup for most people.

To get it running as an executive assistant, I had to:

If you just want a flexible AI assistant that works out-of-the-box, Manus is a better option. Everything comes in one package. No assembly required. No hunting down API keys. No creating OAuth apps in Google Cloud Console. No reading SKILL.md files.

OpenClaw is like building your custom home from scratch. Manus is buying a pre-built home and customizing the cabinets and flooring.

I'll write a full comparison on pros and cons. For now, know this: OpenClaw is powerful, but it's not plug-and-play. If that trade-off doesn't excite you, skip the setup and use something that just works.