Playbook v3.1
Home Our Story Playbook Live
Version 3.1 · April 2026

The Compound Operations Model

Open-source AI operations playbook for consumer SMEs

The complete operational blueprint for building a multi-agent AI system inside a consumer brand — written from 12+ months of running one in production. Every architecture, every config, every lesson learned the hard way, published so operators can fork the repo and adapt it.

Start Reading → Fork the repo
45
Chapters
~5h
Read Time
32
Production Lessons
18:1
ROI

Table of Contents

01 Overview · The thesis, the problem, and the architecture that solves it
01
Introduction
Who built this, why, and what makes it different
6 min → 02
The Problem
Why consumer brands are stuck in operational quicksand
4 min → 03
Architecture
The multi-agent operating system blueprint
6 min →
02 The Brain · The primitive: knowledge, skills, tools — and how teams actually use it
04
The Company Brain
7 min → 05
30-Second Company Brain
5 min → 06
Live Dashboard Tour
4 min → 07
Downloadable Artifacts
5 min → 08
Capability Maturity Ladder
5 min → 09
Memory Architecture
Context Tree, Knowledge Mining, ByteRover native
15 min → 10
MCP Server
44 tools, 167 skills, zero-setup team access
8 min → 11
AI-Native Team Onboarding
Connect every employee to the swarm in 5 minutes
14 min →
03 The Agents · Seven AI agents that run operations on top of the brain
12
Customer Service
CS triage, brand voice, pattern detection at scale
4 min → 13
Inventory & Supply
Multi-location sync, 3PL integration, demand sensing
3 min → 14
Finance & Reporting
Automated P&L, cash flow, variance detection
4 min → 15
Marketing & Lifecycle
Email optimization, segmentation, attribution
5 min → 16
Wholesale & B2B
Account management, order ops, AR recovery
4 min → 17
Retail & Physical
Foot traffic, staffing, store performance
4 min → 18
Merchandising
Sell-through, allocation, pricing, inventory health
5 min → 19
HR & People Ops
Onboarding, vacations, payroll prep, policies
5 min →
04 Results · Honest ROI, production lessons, and advanced capabilities
20
ROI Analysis
Honest 18:1 with every assumption on the table
9 min → 21
Production Lessons
32 lessons learned the hard way in production
17 min → 22
Failure Ledger
5 min → 23
Advanced Capabilities
15 specialized features: AutoResearch, LLM Council, Pattern Library
13 min → 24
Profit Throttle
7 min → 25
Pattern Library
5 min → 26
Knowledge Mining Loop
5 min → 27
Council vs Punta de Flecha
5 min → 28
Invoice Pipeline
4 min → 29
Master Calendar
4 min → 30
Consumer SME Stack Map
3 min →
05 Appendix · Technical implementation depth for operators and engineers
31
Technology Stack
GPT-5.4 via ChatGPT OAuth, Sonnet for CS/HR, hardened infrastructure
9 min → 32
Implementation Paths
Open-source paths: read, fork, local pilot, production adaptation, hands-on help
8 min → 33
OpenClaw Setup Guide
Step-by-step: install, configure, deploy, monitor OpenClaw agents
7 min → 34
EU AI Act Compliance
Risk classification, transparency, HR guardrails, DPIA template
7 min → 35
Brand Bootstrap
One command to go from blank VPS to running swarm for a new brand
11 min → 36
Ingest Layer
How to feed the brain safely — allowlist, DLP, ACL at index boundary, RTBF (Phase 1: structured sources)
11 min → 37
The 5 Pillars
Maps McKinsey 5 pillars of the agentic organization to a practical open-source reference
8 min → 38
Agentic Governance
Three meta-agents watching the seven domain agents: critic, guardrail, compliance
6 min → 39
Agent Factory Pattern
McKinsey 2-5/50-100 ratio — 7 monolithic agents refactored into factories of specialized sub-agents
7 min → 40
LLM Provider Abstraction
5-provider multi-LLM abstraction (Anthropic + OpenAI + Gemini + Qwen + MiniMax). Brand owns its API keys, compAI never touches them.
7 min → 41
Factory Runtime
Smoke test runtime for the Agent Factory. One CLI command dispatches all 10 sub-agents against your configured LLMs and produces a readable trace.
6 min → 42
MVP Runtime
Autonomous runtime MVP — daemon + parallel dispatch + brain lookup + review queue + workflow hooks.
6 min → 43
Webhooks + Slack Digest
HTTP receivers for 4 helpdesks (Richpanel/Gorgias/Zendesk/Intercom) + daily Slack digest. Autonomous end-to-end from ticket arrival to review queue.
6 min → 44
Onboarding Experience
Onboarding pack + setup-brand wizard + team-onboard wrapper — every employee gets the same experience the reference deployment has today, under 30 min per person.
6 min → 45
Source Playbook Index
Source table of contents and version note for the open-source playbook
6 min →