Examples you can copy.

One ChatAgent with state, no UI, no middleware. The smallest possible ayjnt project beyond the blank scaffold — adds persistent messages so you can see how state works on a Durable Object.

Connect from the Cloudflare Agents client SDK. Explains the path vs basePath gotcha and why server-side getAgentByName matters for identity messages.

Root → leaf file-based middleware chain. One `middleware.ts` adds logging + timing headers to every response; a nested one gates a subtree with bearer-token auth. Route groups let you share middleware without nesting URLs.

Two agents talking over Workers RPC via typed `getAgent<T>()`. An Orders agent calls Inventory to decrement stock — full method autocomplete, exception propagation, oversell protection.

Drop an `app.tsx` next to `agent.ts`. The generated `useAgent()` hook is typed to that agent's class and state. Live multi-tab state sync built in — open two tabs, they share state.

Build an MCP server for LLM tool-calling. ayjnt detects `McpAgent` as a base class and dispatches through `McpAgent.serve()` automatically — streamable HTTP and SSE transports handled for you.

An agent that schedules one-shot work with `this.schedule()`. Relative delays (seconds from now), absolute times (ISO dates), and persistent state for both pending and fired reminders.

Agent that wakes itself on a fixed cadence via `scheduleEvery()` and mutates its own state. Classic background-worker pattern on a single Durable Object. Ships with a live bar-chart UI.

Multi-user realtime chat. One DO per room, broadcast via the Agents connection API, co-located React UI with presence + typing indicators — demonstrates the state-sync vs broadcast trade-off.

Streaming chatbot backed by Google Gemini. The DO holds the conversation; the React UI subscribes to state and renders incrementally as setState ticks every SSE chunk. No streaming plumbing in the client.

A two-agent retrieval pipeline. QA agent decomposes the question, fans out retrievals to the Index agent over typed RPC, then composes a grounded answer. Workers AI for both embeddings and generation.

Asteroid-field shooter. One Durable Object owns the world, runs a 30Hz physics tick, and broadcasts the entire state every frame. Clients send keyboard inputs and render the canvas from state.

Two-player chess with turn enforcement + legal-move validation server-side and a React board synced via useAgent. A study in constraining client-side mutations.

A four-agent collaborative system on an asteroid-mining mission. Commander orchestrates navigator, scout, and engineer via typed RPC every 2s. Each role has its own UI so you can dive in and inspect what that crew member knows.