Production AI Agents
Model Context Protocol integrations, Claude Code and Codex Skills, and multi-agent orchestration for NZ and AU enterprises moving beyond AI experiments into production infrastructure.
Close the gap between "Claude can do this"
and Claude doing it in production.
Most organisations discover that the distance between a clever prompt demo and a reliable production agent is larger than expected. We close that gap across three layers.
Architecture
How agents get context, authenticate, call tools, hand off work, and stay observable when things go wrong.
Integration
MCP servers that connect Claude to existing systems. Skills that package domain-specific capabilities. Knowledge graphs that give agents persistent memory.
Orchestration
Scheduled agent tasks, parallel execution, multi-agent coordination, and the patterns that keep complex agent systems legible and maintainable.
How we work
Engagements scale from a one-week review through to full platform build-out.
Architecture Review
We audit your current or planned agent architecture and deliver a written report with specific recommendations.
Proof of Concept
A working implementation of a specific use case, with production-quality code and documentation.
Platform Implementation
Full agent platform build-out for a specific domain or business function.
Private Workshops
For teams exploring where agents fit before committing to a build. On-site or remote.
Capabilities
We work in TypeScript, Bun, and the Anthropic stack. Every pattern we deploy for clients has been battle-tested inside our own operations first.
We run our own agent infrastructure in production every day. The architecture review you hire us for is the same architecture we use ourselves.
- MCP server design and implementation
- Claude Code and Codex Skill architecture and authoring
- Multi-agent orchestration patterns (scheduled agents, parallel execution, context handoff)
- Knowledge graph and RAG integration (pgvector, Neon, custom entity resolution)
- Voice and multimodal AI integration
- Production observability for agent systems
Reference Implementations
What we've built in-house, and what's running for clients.
Mind
Our in-house AI agent infrastructure. Claude Code extended with a library of custom skills, multiple MCP servers, a scheduled agent daemon, voice pipeline, and knowledge graph integration. Demonstrates composable agent architecture in daily production use.
Telepathy Knowledge Graph
Entity-based graph with vector search, edge relationships, and an MCP server exposing full CRUD. The memory layer across multiple AI applications.
Multi-branch Automotive Retailer
Production agent deployment for a NZ used-car retailer. Scheduled Claude Code agents run every day, tailing live auction-site crawls, classifying each run against written procedures, and reporting back. Demonstrates scheduled multi-agent orchestration, live log-tailing from cloud sources, and operational oversight of long-running automated processes for a real commercial client.
Professional Services Deployment
Claude Code with custom domain skills — tax calculations, IRD workflows, reconciliation — deployed into a live NZ accounting practice. Demonstrates the architecture working for regulated-industry, high-accuracy workflows.
Let's talk about your agent roadmap.
Book a no-charge 30-minute discovery call. Bring your questions about architecture, integration, or where to start.
Get in Touch