The Compound Operations Model

An Open-Source AI Operations Playbook for Consumer SMEs

By compAI · Open-source educational portfolio · Built inside a profitable brand growing 50%+ annually · Repo: https://github.com/darLAAGAM/ai-native-playbook (temporary placeholder before the future github.com/ai-native namespace)

---

What This Is

This is not a strategy deck. It's not a trend report. It's not thought leadership.

This is the operational blueprint built — and running — inside a profitable European consumer brand. Eight-figure revenue, healthy EBITDA, 50%+ annual growth, ~40-person team. Seven AI agents in production plus a founder command center — full system cost €352/month, verified value €77K/year in reclaimed labor hours. 18:1 ROI with every assumption on the table. The source is published as an educational portfolio so other consumer SMEs can fork the repo, read the playbook, and adapt the patterns.

Everything in this playbook is real. The architectures are running. The configs are from production systems. The numbers are audited.

The Thesis

For every dollar a brand spends on software, it spends six on services and headcount to operate that software. Shopify costs €2K/year. The people managing inventory, processing orders, answering tickets, and closing the books cost €200K+.

Most AI tools sell another dashboard. This playbook documents the operating layer above the dashboards.

Every improvement in AI models makes the system faster, cheaper, and easier for operators to adapt. We are not positioning this as a SaaS product or a paid repo. We are publishing the working architecture, the lessons, and the artifacts so consumer SMEs can fork the repo, inspect the assumptions, and build their own version.

Who This Is For

Consumer SME operators — beauty, food & beverage, home, wellness, pet, outdoor, fashion, and retail — running on Shopify or a similar commerce stack with €2M–50M in revenue. You have a ~40-person team. Multiple channels. And operations that are creaking under growth.

If you've been told you need to hire 3 more people, this playbook shows you a different path.

The Compound Operations Model™

We've organized this playbook around the open operating model used in the reference deployment:

1. Integrate — Connect your systems into a unified data layer
2. Specialize — Deploy purpose-built AI agents per operational domain
3. Orchestrate — Make agents coordinate across functions
4. Compound — Let the system get smarter every day

Table of Contents

Ch.	Title	What You'll Learn
01	Introduction	Who built this, why, and what makes it different
02	The Problem	Why brands are stuck in operational quicksand
03	Architecture	The multi-agent operating system blueprint
	— The Agents —
04	Agent: Customer Intelligence	CS triage, brand voice, pattern detection
05	Agent: Inventory & Supply Chain	Multi-location sync, 3PL, demand sensing
06	Agent: Finance & Reporting	Automated P&L, cash flow, anomaly detection
07	Agent: Marketing & Lifecycle	Email optimization, segmentation, attribution
08	Agent: Wholesale & B2B	Account management, order ops, pipeline
09a	Agent: Retail & Physical	Foot traffic, staffing, store performance
09b	Agent: Merchandising, Wholesale & Assortment	Sell-through, allocation, pricing, inventory health
	— The Stack —
09c	Agent: HR & People Ops (HR Agent)	Onboarding, vacations, payroll prep, policies, expenses
10a	The Technology Stack	What runs under the hood — models, infra, Claude Code, Agent-Reach, 44 MCP tools
10b	Memory Architecture	Context Tree, Knowledge Mining cron, shared brain sync (SuperMemory deprecated)
10c	The MCP Server	How your entire team gets AI superpowers — 44 tools, zero setup
10d	Advanced Operational Capabilities	15 specialized features that compound over time — AutoResearch, LLM Council, Pattern Library, and more
	— Implementation —
11a	Implementation Paths	Read the playbook → fork the repo → run locally → adapt with human review → ask hello@usecompai.com for hands-on help
11b	Lessons from Production	32 lessons: OAuth failures, memory cleanup, model routing, anti-injection hardening, OpenClaw vs systemd, and more
11c	OpenClaw Runtime Setup	Agent runtime framework — launchd plists, ChatGPT OAuth, cron scheduling
11d	EU AI Act Compliance	Full compliance package — DPIA, AI System Register, Annex III guardrails, Article 50 transparency
11e	Brand Bootstrap (1 cmd)	From blank Ubuntu VPS to running swarm in one terminal command — curl usecompai.com/init \| bash
11f	Ingest Layer (Phase 1)	Feeding the brain safely — allowlist + DLP + ACL at index boundary + RTBF. Structured sources only in v0.4
12	ROI Analysis	Real numbers from 6+ months in production
14	Team Onboarding	Connect every employee to the swarm in 5 minutes — brain access, agents, Google Workspace, zero setup
	— The Future —
15	The 5 Pillars (McKinsey)	The agentic organization framework mapped to compAI — the 1% network model, productized
16	Agentic Governance	Three meta-agents (critic + guardrail + compliance) watching the seven domain agents
17	Agent Factory Pattern	McKinsey 2-5/50-100 ratio: 7 domain agents → factories of 10+ specialized sub-agents each. CS reference shipped v2.6
18	LLM Provider Abstraction	5-provider multi-LLM routing (Anthropic + OpenAI + Gemini + Qwen + MiniMax). Brand-owned API keys. Fallback chains. Per-sub-agent routing
19	Factory Runtime v0.9.0	operai-init factory run-once — smoke test: 10 sub-agents dispatch end-to-end, full markdown trace, mock-LLM mode
20	MVP Runtime	Autonomous daemon: events → parallel dispatch → review queue. Workflow hook points for brand-specific extensions. Honest 70% of the reference swarm
21	Webhooks + Slack Digest	HMAC-verified receivers for 4 helpdesks + daily Slack digest. Autonomous end-to-end from customer email to review queue
22	The Onboarding Experience	Open onboarding pack (skills + custom instruction + templates) + setup wizard + team-onboard wrapper. 30 min per employee, same experience as the reference deployment

How to Read This

• Founders/CEOs: Read Ch. 1–3 and Ch. 12. That gives you the thesis, the architecture, and the business case.
• Ops/Tech Leads: Read everything. Ch. 4–9 are your implementation guides. Ch. 11b is the production war stories that will save you days of debugging.
• Investors/Board: Ch. 12 has the numbers you want. Ch. 3 has the architecture. Ch. 10c shows the team-wide impact.

---

This playbook is a living document. The system it describes is in production and evolving daily. When we learn something new, we update the playbook.

Version 2.0 · 12 April 2026 — EU AI Act fully compliant (24/24 items, DPIA + AI System Register documented), honest 18:1 ROI with auditable math (€77K value / €4.2K cost), 32 production lessons, 15+ advanced capabilities (Punta de Flecha adversarial cross-model deliberation, AutoResearch, LLM Council, Pattern Library live), ChatGPT OAuth model strategy (5 agents at €0 incremental), anti-prompt-injection hardening across all SOULs, ACK rule fleet-wide, HR guardrails against Annex III high-risk uses, CS Article 50 transparency disclaimer deployed. Live at usecompai.com with public dashboard and free playbook. Fork the repo at https://github.com/darLAAGAM/ai-native-playbook, or contact hello@usecompai.com for hands-on implementation help.