Skip to content

AI Agent Infrastructure & Management Portal

An Australian events and marketing company

OpenClaw Rails 8 Falcon SQLite Claude API Mac Mini M4 Tailscale Google Workspace APIs Discord Monday.com Xero

The Challenge

An Australian events and marketing company with over 15 years in the industry wanted to use AI agents to automate key business functions — integrating with platforms including Xero (accounting), ticketing platforms, Meta Ads (advertising), fan data platforms, and Monday.com (project management).

The company needed three things: specialised AI agents designed and built for their specific business operations, a proper security architecture that isolated sensitive financial agents from staff-facing ones, and a management portal that gave the whole team visibility and control over their AI workforce.

The managing director also needed to bring on staff to help manage agents, but OpenClaw's security model is explicitly single-operator — it doesn't support multi-tenant isolation. A staff member with Gateway access could theoretically read API keys and financial data belonging to a different agent.

The Solution

I delivered this engagement in two phases. First, the agent infrastructure — three specialised agents across a security-isolated dual-Gateway architecture with Google Workspace integration, Discord channels, and a structured knowledge vault. Second, a custom Rails 8 workspace portal built in 3.5 days that gives the team a unified interface to manage all three agents.

Phase 1: Agent Infrastructure

I designed and built three specialised agents, each with distinct roles and scoped access:

Rock — the primary business operations agent, integrated with Xero accounting, ticketing platforms, Meta Ads, fan data platforms, and Monday.com project management.

Zed — the managing director's personal assistant, handling email, calendar, private admin, and tasks.

Zaphod — the staff-facing agent covering marketing (social media, content, campaigns), sales (ticket sales, client relationships, promo codes), and venue operations (logistics, supplier coordination).

Two-Gateway Security Architecture — rather than building a cosmetic permissions layer on top of a fundamentally single-tenant system, I designed real isolation using the operating system itself. Gateway A runs under the owner's macOS user account and hosts Rock and Zed — the agents with access to financial data and personal information. Gateway B runs under a separate macOS service account and hosts Zaphod, the staff-facing agent. The OS enforces the boundary: the staff process physically cannot read files under the owner's home directory.

Discord Integration — the team communicates with agents through Discord, with topic-specific channels for different workflows.

Google Workspace Integration — agents operate through the company's Google Workspace via domain-wide delegation for Gmail, Calendar, Drive, and Docs.

Structured Knowledge Vault — an Obsidian-based knowledge vault as the company's institutional memory covering venue profiles, talent roster, campaign playbooks, competitor intelligence, and operational procedures.

Phase 2: The Workspace Portal

Unified Agent Dashboard — all three agents visible at a glance with Gateway assignment, online status, skill count, and quick stats.

Skill Management — a searchable skill browser with editing capabilities, approval workflows for non-owner edits, and cross-agent skill duplication.

Document Portal — dual-source document browser for both the shared knowledge vault and per-agent workspace files.

Task Board — Kanban board with agent assignment and real-time streaming updates via Turbo Streams.

Agent Chat — real-time chat interface connected to selected agent's Gateway via ActionCable.

Role-Based Access Control — owner sees all agents across both Gateways and can edit directly; managers see only assigned agents and submit edits for approval; staff can view but not modify.

Phase 3: Accounts Receivable Automation (In Design)

The engagement expanded into traditional business process automation — designing a complete AR system connecting Monday.com, Xero, and Google Cloud Functions to close a revenue leakage gap where events were being missed for invoicing.

The Results

Three specialised AI agents designed and built — Rock (business operations), Zed (personal assistant), and Zaphod (marketing/sales/ops) — each with scoped access to relevant business tools and its own personality, memory, and skills.

Real OS-level security isolation — sensitive credentials and financial data protected by macOS filesystem permissions, not just application logic.

Custom workspace portal — built in 3.5 days, providing unified agent management, skill editing, document access, task coordination, and real-time chat.

Seven production skills deployed — integrations with accounting, ticketing, advertising, project management, and messaging platforms.

Structured knowledge vault — institutional intelligence captured and accessible to all agents.

Staff onboarding ready — role-based access control enabling the managing director to delegate agent management without exposing sensitive data.

AR automation designed — end-to-end accounts receivable system from event logging through to invoicing and weekly reconciliation.

Zero-ops deployment — single Mac Mini M4 deployment with launchd and Tailscale, no cloud infrastructure.

Remotely managed — ongoing development and optimisation handled from the Gold Coast without on-site visits.

~60 hours of consulting delivered across agent infrastructure, portal development, automation design, and advisory.

Want Similar Results?

Let's talk about how I can help your business.