Merlin

An assistant that pays attention — even when you're not asking.

Most assistants wait. You type, they reply, the moment passes. Merlin stays on. It learns what you like, watches what matters, and quietly handles the things you'd rather not think about.

Runs on the Claude subscription you already have. No API keys. No per-token bills.

On the Claude you already have

You may have seen agent frameworks like openclaw. They're powerful, but they plug into the Claude API — separate account, pay per token, a bill that scales with how much your agent does.

Merlin works the other way around. It runs through Anthropic's official Claude environment on the subscription you already use in your browser. Sign in once, and Merlin works through your account.

If you'd rather pay by usage, Merlin will use an API key too. But the subscription path is the one it was designed for.

What Merlin does

What makes Merlin different

Five things most AI assistants miss.

🧠 It remembers you

Your preferences, your decisions, the things you said last month — Merlin keeps them. The longer you use it, the more it understands what you actually want, not what most people might want.

🐝 It reaches out

When Merlin notices something worth knowing — a flight delay, a calendar conflict, a chance to catch a show you'd love — it tells you. Quietly. At the right moment. When there's nothing useful to say, it stays quiet. (No spam. No "did you know…" pings. Just signal.)

🛠️ It gets things done

It doesn't just suggest. It writes the email, books the dinner, sets the reminder, runs the research. With safety rails for the things that matter — no money moves, no commitments made on your behalf — without you saying yes first.

It shows up on time

Recurring work happens on schedule, in the background. Morning briefs, weekly recaps, the things you'd otherwise have to remember. You don't have to. Merlin does.

📈 It gets better as you live with it

Every week, Merlin reviews how it did. Were its messages useful? Did they land at the right time? It learns from how you respond — promotes what works, drops what doesn't, and gradually shapes itself to fit your life.

Why we built it

A real assistant remembers what matters, notices what's worth noticing, takes action when it can, and learns as it goes.

Chatbots do none of those. Merlin does all four — and that's the difference between a search box that can talk and an assistant that can actually help.

It runs on your hardware. It works with your accounts. The integrations are deliberately spare — you wire your own email, calendar, smart home — because your assistant should know about your life, not someone else's.

github.com/mraccah2/merlin →
Open source · Apache 2.0 · Yours to shape

Under the hood

For the curious. Merlin is built on Claude Code with persistent claude -p supervisors, durable session resume, an MCP shell bridge that turns any CLI into an agent tool, a SQLite memory layer with full-text + embedding search, a local Ollama model handling latency-sensitive acknowledgments, and markdown playbooks describing every scheduled job. Read the architecture: github.com/mraccah2/merlin/blob/main/system/architecture.md.