OpenClaw System Architecture (Public Overview)

Last updated: 2026-03-05

Executive summary

We run an AI-agent operating system (OpenClaw) that lets a small team of specialized agents execute real work across marketing, product, engineering, and ops.

This page is designed to be shareable with viewers:

What this document is / isn’t

This is: a practical blueprint showing how we structure and run an agent-driven org (interfaces, routing, knowledge, tools, and delivery).

This is not: a dump of secrets. We intentionally do not publish:

High-level architecture

     (Human)
      Shad
       |
       | Slack DM / channel message
       v
+--------------------+          +--------------------+
| Communication      |          | Knowledge / Source |
| Slack              |<-------->| of truth           |
+--------------------+          | Notion             |
       |                        +--------------------+
       | events/messages
       v
+--------------------+
| Orchestrator        |
| OpenClaw Gateway     |
| - routes messages    |
| - runs agents        |
| - tool allowlists    |
+--------------------+
       |
       | delegates work
       v
+------------------------------+
| Agent layer                   |
| - Chief of Staff (router)     |
| - Division heads (SEO/Product |
|   /RevOps/Eng/Design)         |
| - Specialists (when needed)   |
+------------------------------+
       |
       | tool calls (Notion, GitHub, analytics, etc.)
       v
+--------------------+
| Execution systems   |
| (3rd party + infra) |
+--------------------+
       |
       | deliverables + status
       v
  Notion pages + Slack updates

Core components (with responsibilities)

1) Communication layer (Slack)

Role: fast human-in-the-loop control interface.

What happens here: