Agent Patterns

Runnable patterns for building with agents.

Each pattern (Pipeline, Decomposition, Plan → Execute, Debate, Reflection, Routing, Best-of-N, Handoff, ReAct, Human-in-the-Loop) is defined as a JSON file describing its agent topology. Open one to inspect the structure, send it a prompt, and watch the agents work.

Stack

• Next.js 16 (App Router)
• Vercel AI SDK + @ai-sdk/anthropic
• Tailwind CSS
• All patterns run on Claude Haiku 4.5

Run locally

npm install
echo "ANTHROPIC_API_KEY=sk-ant-..." > .env.local
npm run dev

Open http://localhost:3000.

How patterns work

Each pattern is a JSON file in engine/patterns/ describing:

• nodes — agent definitions (role, system prompt, inputs, outputs)
• execution — a tree of step types (sequence, parallel, map, loop, branch, dynamic, tool_loop, hitl_checkpoint)

The interpreter in engine/interpreter.ts walks that tree and calls Claude. The same JSON drives the topology visualization, the gallery's auto-derived tags, and the fake animated previews on the home page.

To add a new pattern: create a JSON file, register it in engine/index.ts and lib/pattern-meta.ts. Both the runtime and the UI pick it up for free.