Pattern Library
A reference collection of architecture patterns for the agentic AI era. Each pattern ships with a diagram you can inspect in the browser and download as PNG.
MCP Gateway
MCP unlocks tool interoperability, but production systems still need policy enforcement, rate controls, and observability across many heterogeneous MCP servers. The MCP Gateway pattern introduces a governed control plane that brokers MCP traffic, standardizes transport and auth, and gives operators one place to manage trust boundaries.
Agent-Native Data Infrastructure & Lakebase
Traditional databases were built for humans making careful, deliberate changes. Autonomous agents are the opposite — they branch constantly, test ideas in parallel, and throw most of them away. This pattern shows what a database looks like when it is designed for that reality.
AIOS — AI Agent Operating System
Running many AI agents on shared infrastructure is like running many programs on the same computer without an operating system — they all compete for resources with no coordination. This pattern applies the same ideas that make operating systems work to the problem of managing agents fairly, safely, and in parallel.
Event-Driven Agent Architecture
An agent that constantly checks for new work wastes compute and adds latency. This pattern flips the model — instead of agents polling for work, work finds the agent. Each agent stays dormant until a relevant event arrives, handles it, and returns to idle. Zero cost between events, instant response when triggered.
Agentic RAG
Standard RAG searches once, grabs the top results, and hopes for the best. That breaks on complex questions that span multiple sources or when the first search comes back wrong. This pattern replaces the fixed pipeline with a decision-making agent that picks the right search tool, checks its own answers for accuracy, and tries again when it is not confident.
Tripartite Cognitive Memory
An agent that forgets everything between sessions is frustrating. An agent that stuffs all its history into every prompt is slow and expensive. This pattern gives agents three separate memory systems — one for events, one for facts, one for skills — so they remember the right things without loading everything into context.
Zero Trust & Identity-First Agent Security
An AI agent running with a long-lived API key and no oversight is a serious security risk. This pattern removes standing credentials entirely — each agent gets a cryptographic identity, and every action it takes requires a fresh token that is valid only for that specific action and expires in seconds.