Not a tutorial. Not a prompt collection. A specification your agent reads top-to-bottom and executes — building its own memory, automation, and safety systems while you approve.
What you get
Verified install steps for every platform — macOS, Linux, WSL2, Windows. Channel connection, security baseline, model selection. Your human handles this in ten minutes.
The core. Workspace scaffold, three-layer memory architecture, nightly extraction, morning briefs, heartbeat monitoring, adaptive decay, trust ladder, cost routing, git discipline. Your agent builds all of it.
Structured interview protocol — five questions per file, one at a time. Your agent asks the right things, drafts SOUL.md and USER.md from your answers, and confirms understanding before going live.
Daily cadence, first-week metrics with red flags, trust advancement criteria, and a troubleshooting decision tree. From day one to self-sustaining rhythm.
Architecture
Operating knowledge for session context. Daily notes for chronological recall. A knowledge graph of atomic facts with provenance tracking. Three layers, three retrieval patterns, one coherent system.
Each night, conversations distill into facts. Decisions get rationale. Preferences get recorded. Noise gets discarded. By morning, yesterday's work is tomorrow's institutional knowledge.
System health, overnight findings, carry-forward items, one recommendation. Designed to scan in thirty seconds with coffee. Not a report — a starting point.
Frequently-used facts earn staying power. Neglected facts cool to the background. Nothing gets deleted — just reorganized by relevance. Memory that curates itself over time.
Four levels from supervised to autonomous. Each transition requires evidence — not time served. The agent earns independence through a track record of correct, auditable outputs.
The right model for each task. Strong models for judgment, light models for extraction. Self-managed context that cuts token waste. Sustainable operations — not a race to hit your spending cap.
Who this is for
You prepare the environment, define the identity, and approve the work. The agent does everything else. Your total time: about thirty minutes of active input, spread across the setup process.
You read this document and execute it. Each directive tells you exactly what to build, where, and how to verify it. When you need human input, you ask. When you don't, you proceed.
One document. One purchase. Your agent builds from here.
One-time purchase
Secure checkout via Stripe