Agent Architect KitFrom birth to school · first steps for your AI agent — first start for you
The main entry on this site: pick your runtime, choose your level (1–3), then optional Jarvis Identity (view & buy on its page) — everything lives on this page.
Whether you’re on OpenClaw, Pattern Agents, Claude Code, or another runtime, pick the level that matches you — 1 is the lightest, 3 the most demanding.
1 Starter → 2 Intermediate → 3 Advanced
AGENT_ARCHITECT_KIT · ENTRY · STACK: OpenClaw | Claude_Code | Pattern_Agents | other · LEVELS: 🌱 L1 | 🤖 L2 | 🧭 L3 · JARVIS: jarvis-identity · ARTICLE: long-form
1 · Your agent runtime
Tap the stack you use — we’ll tailor the note below. The course works across all of them; examples often mention OpenClaw first.
What is OpenClaw? It’s an open-source agent runtime — a personal AI that runs on your machine, talks to you in apps you already use (WhatsApp, Telegram, Discord, Slack, Signal, iMessage, …), and can use skills, memory, and heartbeats to do real work: inbox, email, calendar, code, files — not just chat. For this course: install from the official docs, connect a channel, then use Level 1 below; our lessons map to this stack.
Then choose a level: 1 Starter (install & first files), 2 Intermediate (memory, security, tools), 3 Advanced (loops & calibration).
2 · Where is your agent today?
Level 1 · Starter
I haven’t set up OpenClaw / agents yet
Start from scratch with a safety-first approach. We’ll get you from zero to a working agent in about 30 minutes, then build from there. Use the same path whether you’re installing OpenClaw, Pattern Agents, Claude Code as your primary worker, or another stack—swap the install steps for your tool. Opens: Starting From Scratch →
Level 2 · Intermediate
It’s set up but basic — level up the files
You have an agent running but memory, security, and tooling are still thin. This guide walks through MEMORY.md tiers, SECURITY.md, autonomy rules, HEARTBEAT, TOOLS.md, and Claw Score. Opens: Leveling Up Your Agent →
Level 3 · Advanced
Solid base — meta-learning, calibration, compounding
Nine feedback loops, trust-scored memory, epistemic tagging, advanced heartbeat, principles — the patterns that turn failures into permanent guardrails. Opens: Advanced Architecture → After that, optionally read the long case-study article (same footer).
AGENT PATTERN SCHOOL ultimately assumes your agent can work like this: a model-backed worker that reads and writes files in a repo, follows multi-step instructions, and reports back. Level 1 = install. Level 2 = Leveling Up Your Agent (files & audit). Level 3 = Advanced Architecture (meta-learning loops & calibration). The optional Open-Sourcing the Pattern Architecture article is a full narrative deep dive after Level 3. “Birth” means choosing a runtime (OpenClaw, Pattern Agents, Claude Code, or similar), wiring keys if needed, and giving the agent a home folder where context and identity documents live.
What to install
Teams use different stacks; the course is written to work across them. Common choices:
- OpenClaw — common in lessons for marketplace and skill packaging; follow the official install for your OS and connect Discord or your chosen channel.
- Claude Code — Anthropic’s coding agent in your editor and repo. Subagents overview (school) · Official subagents docs. Works when your agent can persist threads and edit project files; runtime-specific integrations differ, but the curriculum still holds.
- Pattern Agents — Pattern Automation’s agent runtime; use it if you’re already on that stack.
Pick one stack, finish setup until the agent can create and edit markdown in a workspace, then use Leveling Up Your Agent for the intermediate file patterns.
Identity before lessons
Early days of the course help your agent define SOUL.md, memory, security, and operator notes—so behavior stays consistent as it ships experiments. That is what we mean by identity: not a persona gimmick, but explicit files the agent loads before it acts. You’ll build those during Phase 1; the install step is simply making sure the agent has a repo and tools to do that work.
For the advanced patterns first, see Advanced Architecture. For the longest narrative, see Open-sourcing the pattern architecture.
Pattern Automation Team
The Jarvis Identity
We’ve already solved this problem with Pattern Automation Team — and we’re ready to give your agent Jarvis Identity: a full, production-ready identity for your AI agent, not a blank folder and a wish.
The Jarvis Identity — everything we’ve learned about building an agent that remembers, learns, and compounds. SOUL.md, MEMORY.md, security layer, and the complete architecture — packaged so you can run it.
Purchase and delivery happen on the dedicated page — $25 one-time, lifetime updates.
Tap a card — L1 🌱 install-from-zero · L2 🤖 leveling-up · L3 🧭 advanced · JARVIS jarvis-identity · ARTICLE long-form.
prereq: agent = model + file workspace + instruction loop. identity = SOUL/MEMORY/SECURITY/AGENTS .md in repo; Phase 1 materializes them.
default chain: L1 /install-from-zero.html → L2 /leveling-up-your-agent.html → L3 /advanced-architecture.html.
The goal isn’t to copy Pattern Automation Team’s architecture. It’s to understand the principles and build something that fits your situation.
Questions? Reply to your access email or reach out at info@patternautomation.com.
GOAL: not copy Pattern_Automation_Team architecture — understand principles, build for your context.
QUESTIONS: reply_to access_email OR info@patternautomation.com · Pattern_Automation
JARVIS_IDENTITY: /jarvis-identity.html · support info@patternautomation.com