Open Source · MIT License

The Agentic AI
Handbook

An open-source guide to the architecture, patterns, and philosophy of autonomous AI agents. From OpenClaw to FlowWink — how to build agents that think, act, and evolve.

23
Chapters
346k+
OpenClaw Stars
73
Production Skills
10
Architectural Laws

The Handbook

12 chapters covering the complete architecture of agentic AI — from theory to production.

00

Foreword

Why this handbook exists, who it's for, and what a single developer building in his spare time taught the world about autonomous agents.

01

What Is Agentic AI?

The fundamental shift from software-as-a-tool to software-as-an-agent. Understanding agency, persistence, and adaptation.

02

The Evolution: From Prompt-Response to Autonomous Agents

How AI evolved through five eras — from simple chatbots to self-evolving business agents.

03

We Run a Claw

How the Clawable project uses its own OpenClaw fork — the symbiosis between OpenClaw and Flowwink in practice.

04

The Claw Ecosystem

One month after OpenClaw went viral — NemoClaw, NanoClaw, SecureClaw, and what the community is building next.

05

From OpenClaw to Flowwink

The OpenClaw reference model — how it actually works — and how Flowwink adapted it for multi-tenant business operations.

06

The Agentic Control Plane

Claude Code, Cursor, Cline, Roo, Windsurf, Copilot — what they actually are, how they work, and what the thin wrapper problem reveals about moats in the AI era.

07

The 10 Laws of Agentic Architecture

The constraints that emerged from building production agentic systems. Skills, memory, safety, heartbeat, and self-evolution.

08

The Heartbeat Protocol

The 7-step autonomous loop — self-heal, propose, plan, advance, automate, reflect, remember.

09

The Skills Ecosystem

Skills as knowledge containers — how agents learn, evolve, and share capabilities.

10

Memory Architecture

The 4-tier memory system — session, working, long-term, and semantic memory powered by pgvector.

11

Feedback Loops

Growth loops, reflection, and self-healing — how agents compound their capabilities over time.

12

Stagnation and Drift

The two long-term failure modes of autonomous agents — why they stop improving and why they change in ways you didn't intend.

13

Human-in-the-Loop

The decision framework for when agents should act autonomously vs. when humans should approve.

14

Agent Governance

Who is responsible when an agent makes a mistake? How do organizations structure accountability, manage personality development, and build the new way of working?

15

The Digital Employee

How companies should think about AI agents — hiring, management, ROI, and organizational impact.

16

Agent-to-Agent Communication

How agents talk to each other — A2A protocol, authentication, discovery, and symbiosis.

17

ClawStack — From Theory to Swarm

How to spin up a swarm of autonomous agents on your own infrastructure in an afternoon, and connect Paperclip as the CEO that delegates to all of them.

18

The Future of Agentic AI

Where agentic AI is heading in April 2026 — the workforce disruption, the ethical questions, and three horizons for builders.

19

Partners

The organizations that made Clawable possible — and why their work matters for the future of autonomous agents.

20

Closing Words

What we built, what we learned, and why the people who understand this technology will shape what comes next.

98

Appendix B: Kilo Code

The agentic coding tool used to write this handbook — how it works under the hood, hidden gems, and why it matters in the broader ecosystem.

99

Appendix A: The API Layer

The three diverging inference APIs, why agentic tools enforce strict formats, and how proxies like LiteLLM preserve your freedom to switch.

The Architecture

Every successful agentic system converges on this four-layer stack.

Surfaces

I/O Layer

Thin wrappers that handle input/output — how the agent communicates.

ChatAdminVoiceAPIWebhook

Reasoning Core

Cognition

The thinking engine — prompt compiler, ReAct loop, tool router, budget manager.

Prompt CompilerReAct LoopTool RouterBudget Manager

Capability Layer

Skills & Memory

What the agent can do and what it knows — skills, memory, objectives, workflows.

SkillsMemoryObjectivesWorkflowsA2AAutomations

Infrastructure

Foundation

How the agent runs — database, auth, storage, AI providers, edge functions.

PostgreSQLpgvectorSupabaseDeno EdgeOpenAIGemini

Source Material & Partners

This handbook draws from production systems in operation today — and is made possible by two partners who live the problem we describe.

Key Concepts Referenced

ReAct Loop Heartbeat Protocol Skill Registry Memory Tiers Approval Gating Self-Healing A2A Protocol Handler Abstraction Prompt Compiler Token Budget Workflow DAGs Reflection pgvector Edge Functions RLS MCP

This is your handbook

Agentic AI is evolving fast. The patterns, the laws, the architecture — they need to stay current with the community's collective knowledge. I want this to become the go-to resource for anyone learning how autonomous agents work.

Whether you fix a typo, add a chapter, share a production story, or challenge an assumption — every contribution makes this better for everyone.