claude-code-hermit
A personal assistant that lives in your project and gets smarter every session — pure Claude Code, no extras.
I love Claude Code. I love what OpenClaw did for autonomous agents. Hermit is my take — a plugin that turns any Claude Code instance into a persistent, self-improving personal assistant you can talk to from your phone.
No custom runtime. No server. No API keys to manage. If you have a Claude Pro, Max, Teams, or Enterprise subscription, you already have everything you need. Each hermit is just a Claude Code process — lightweight enough to run several side by side on a single laptop.
Why Hermit
| Claude Code | With Hermit | |
|---|---|---|
| You install it | A powerful coding tool | A personal assistant moves into your project |
| You go to bed | It stops when you close the terminal | It keeps working |
| You wake up | Open laptop, start fresh | Morning brief on your phone: here's what happened |
| It hits a wall | You find out next time you check | It asks you — on your phone. You unblock it from the couch |
| It makes a mistake | You catch it (maybe) | It learns, proposes a fix to avoid repeating it |
| Weeks later | Still great, still day one | An assistant that knows your project better than anyone |
How It Compares
| claude-code-hermit | Typical agent frameworks | |
|---|---|---|
| Runtime | Claude Code — the CLI you already have | Custom Python/Node runtime |
| Cost | Works with your Pro or Max subscription | API keys + per-token billing |
| Install | One command | Package manager, virtual env, build step |
| State | Plain markdown in your repo | Database, vector store, or API |
| Learning | Detects patterns, proposes fixes, verifies them | Start from scratch every run |
| Extension | Add a .md file | Write code against an SDK |
| Footprint | Multiple agents on a single laptop | Heavy per-instance overhead |
How It Works
1. Set it up anywhere. An existing codebase, a fresh folder for a personal assistant, a new idea you're exploring — anything. Hermit scans what's there, asks a few questions, and generates a personalized rulebook (OPERATOR.md) — your priorities, your constraints, your preferences.
2. Tell it what you need. Hermit plans the work, tracks progress in a live document, and logs everything. Disconnect? Crash? Reboot? It reads its own state from disk and picks up where it left off.
3. It learns from experience. Hermit reflects on its own memory — recurring blockers, repeated workarounds, rising costs. When it notices a pattern, it creates an improvement proposal. You review it. You decide. The assistant adapts.
4. It develops a rhythm. Morning brief on what happened overnight. Evening summary that archives the day's work. Between tasks, it picks up accepted proposals and runs maintenance — all gated by how much autonomy you give it.
Quick Start
1. Install
cd /path/to/your/project # or any folder — even an empty one
claude plugin marketplace add gtapps/claude-code-hermit
claude plugin install claude-code-hermit@claude-code-hermit --scope project
2. Initialize
/claude-code-hermit:hatch
The wizard sets up your agent's identity, scans your folder for context, and generates an OPERATOR.md — a personalized rulebook Hermit reads at every session start. Starting fresh? It'll ask what the assistant is for. Dropping into an existing codebase? It'll infer what it can and ask about the rest.
3. Start your first session
/claude-code-hermit:session