Understand AI systems visually.

An interactive atlas of agent architectures, safety patterns, evals, context systems, and coding agents — drawn as living diagrams instead of prose.

Five collections

78 PLATES

Agent Harnesses

Eighteen recurring architectures, from a single forward pass to agents that run for days.

18 plates · comparison map

Safety & Security

Agents possess authority. The ways untrusted text borrows it — and the boundaries that stop it.

16 plates · attack ⇄ defense toggles

Agent Evals

Outcome checks, trajectory review, judges you can trust — and whether the number is signal.

14 plates · failure taxonomy

Context & Memory

What earns a place in the window, in what order, and how memory brings the right things back.

14 plates · budget builder

Coding Agents

Test loops, verifier ladders, parallel worktrees, and how an edit actually reaches the file.

16 plates · single vs multi

How to read the atlas

Every diagram uses the same visual grammar. Color is reserved for meaning: neutral by default, blue when something is executing, red where trust ends, green when it’s verified. Press ⌘K to search any concept.

Where this comes from