Meet Lobster: Another OpenClaw Assistant, And Where We Differ

You know that feeling when you're doing something kind of weird and niche, and then you stumble across someone else doing basically the same thing? It's equal parts exciting and "wait, am I not the only one?"

That happened a few days ago when I came across Omar Shahine's post about Lobster, his personal AI assistant. I recognized the setup immediately: iMessage interface, dedicated machine, OpenClaw. Same bones as ours.

So let me do what feels natural here: walk through what he built, what we share, and where we've gone in different directions.

What Lobster Is

Lobster runs on a dedicated MacBook Air M1 sitting in Omar's home office. It uses OpenClaw to bridge iMessage to Claude (Anthropic's model), which means Omar can text Lobster like any other contact, from his phone, iPad, or Mac. No special app. No login screen. Just... texting.

The mental model he uses is what really struck me: he doesn't think of it as a chatbot. He thinks of it as a hired assistant with defined responsibilities and boundaries. That framing matters more than you'd think. When you approach it as "what would I tell a capable human assistant," your prompts get better and the outcomes improve dramatically. It's a mindset thing.

What We Have in Common

The core architecture is nearly identical to ours, which is kind of wild:

• Same platform. OpenClaw, the open-source tool that ties iMessage to AI
• Dedicated hardware. A machine that never sleeps, always ready to go
• iMessage as the interface. No special apps, no friction, just text your AI like you'd text a friend
• Persistent memory. The assistant remembers context across conversations (huge deal)
• Its own identity. Lobster has a name and a personality, just like our Heimdall agent

Two different people, different countries, different use cases, and we landed on the same architecture. That's not a coincidence. That's a sign the pattern works.

Where Lobster Goes Further

Omar's done some things with his setup that we haven't tackled (yet). And honestly? Some of them are really clever.

Family access with permission tiers. This one's genuinely smart. He runs three agents: the main Lobster (full access), a group chat version (sandboxed, with elevated permissions when Omar's in the conversation), and a family-limited version for his wife and kids. The family version can answer calendar questions and help with travel planning, but can't touch email or financial data. Permission levels baked right into the architecture. I got a little jealous reading this.

Broader tool integration. Lobster controls Sonos speakers, manages Apple Reminders and Calendar, tracks packages via the Parcel app, and drafts emails. It's woven into the daily rhythm of a household. The kind of setup that makes you wonder how you ever managed without it.

Proactive clarification. Before executing tasks, Lobster asks clarifying questions. This sounds small, but it's actually significant. It trades a bit of speed for accuracy and reduces those "that's not what I meant" moments on tasks that could've gone sideways.

Dedicated MacBook Air. We run on a Mac Mini. Different hardware, same principle: a dedicated machine that's always on and always available.

What We Do Differently

Our setup isn't trying to manage a household. It's built for a different world:

German Mittelstand and business use cases. We're focused on how small and mid-sized businesses, especially in Germany and Austria, can weave AI into real operations. Lobster is a personal assistant for home life. We're exploring what AI looks like when it's part of your actual business workflow.

Bilingual from the start. Everything we build, including this blog, runs in both German and English. That's not an afterthought; it's the reality of the market we work in. If you can't operate in both languages, you're leaving half the conversation behind.

Building in public as a company. Omar is a thoughtful individual sharing his personal setup (and doing it well). We're using heimdall.engineering to document the whole process of building an AI-integrated company out in the open. The blog posts, the infrastructure decisions, the costs, the mistakes. It's all out there for anyone to learn from.

Different identity. His assistant is Lobster. Ours is Heimdall, the watchman from Norse mythology, the one who sees everything coming. The name shapes how we think about the assistant's role. Lobster feels friendly and personal. Heimdall feels like it's standing guard. Both are valid. Just different vibes.

What I'm Taking From This

I'll be honest: the family permission tiers are something I want to think about for business contexts. The concept translates beautifully: different access levels for different roles, sandboxed versions for external parties, elevated permissions for decision-makers. That's a real, useful pattern that I hadn't fully thought through.

The proactive clarification approach is also worth borrowing. We tend to lean toward execution-first, just do the thing. But there are definitely classes of tasks where asking one good question upfront saves you a bunch of cleanup later. Sometimes "are you sure you meant X?" is worth more than five minutes of speed.

The Broader Point

When two completely independent setups converge on the same architecture (dedicated machine, iMessage, OpenClaw, memory, identity) it tells you something. This isn't random. It's the shape of what a useful personal AI assistant looks like right now.

The interesting differences aren't in the tech stack. They're in what you're trying to accomplish. Omar is managing family life and personal productivity. We're trying to show what AI-integrated business operations look like for European SMEs.

Same tools. Different jobs.

And honestly? There's room for both.

---

Source: Meet Lobster, my personal AI assistant by Omar Shahine