CodeCrew: Sovereign AI Infrastructure

The Manifesto: End the AI "Black Box"

You don't need a containerized bubble. You don't need a subscription. You don't need to guess if your AI is phoning home or why it "forgot" your project requirements.

CodeCrew is a paradigm shift. We have moved away from the "App as a Container" model and toward a "System-as-Filesystem" architecture. Your AI isn't an application; it is a permanent, transparent, and hackable layer of your operating system.

---

🔱 The Three Pillars

• CodeCREW — The Cockpit (TUI) A lightweight, high-performance interface for real-time interaction. Built to stay out of your way and focused on the work.

• CodeMAID — The Director (Dashboard) Your management plane. FastAPI-backed, HTMX-powered. It doesn't just manage settings—it governs the integrity of the agent ecosystem.

• CodeMOP — The Engine (Manager Of Personas) The heart of the beast. It provides Long-Term Memory (SQL-backed decisions), Short-Term Context (KV-Buffer), and Directory-Aware Persona Injection (rtfm.md).

---

🧠 Why CodeCrew is Different

1. Filesystem-as-API

We don't hide your AI’s "brain" inside an opaque Docker container.

• Skills are scripts: Want a new AI skill? Drop a script in ~/.agents/tools/.
• Memory is persistent: Decisions are markdown files. You can grep your AI’s history. You can edit its memory. You can physically prune its decision-tree.
• Context is structural: We don't "prompt" the AI; we provide rtfm.md files that define your project's reality.

2. Radical Transparency

The AI is under your supervision.

• Thought Monitoring: A real-time thought.log gives you a window into the agent's intent before it touches your disk.
• Action Guard: No process is executed without an audit trail. You are the architect; the AI is the apprentice.

3. Hardware Sovereignty

Designed specifically for local dual-GPU rigs. It budgets your VRAM as a smart, self-managed resource—dynamically shifting between "Lean" and "Fat" models based on project complexity, not just hardware limits.

---

🚀 The Local-First Workflow

1. Initialize: codemop-setup audits your hardware and optimizes your agent environment.
2. Work: Move into any project folder. The AI "discovers" the agent it needs based on the rtfm.md symlink.
3. Govern: Use the Dashboard to review decisions, prune memory, and audit the agent's research activity.

---

The AI is not a service. It is a local intelligence, rooted in your filesystem, governed by your truth.

---

📜 License

MIT — Michael Robinson (RuMoR / CuckaOccurs)