arcie.
++

Quickstart

From an empty directory to a running, tool-using agent in a few minutes.

This guide takes you from zero to a streaming, tool-using agent.

1. Create the project

npx arcie@latest init my-agent
cd my-agent
npm install

2. Set your Cencori key

export CENCORI_API_KEY=csk_...
export CENCORI_PROJECT_ID=proj_...

3. Define the agent

agent/agent.ts declares the model and Cencori configuration:

import { defineAgent } from "arcie";
 
export default defineAgent({
  model: "claude-sonnet-4-5",
  name: "my-agent",
  cencori: {
    project: process.env.CENCORI_PROJECT_ID,
    billing: { budget: "50.00/month" },
  },
});

Write the system prompt in agent/instructions.md:

You are a helpful AI agent built with Zett on Cencori.
Be concise, accurate, and use your tools when appropriate.

4. Run the dev server

npm run dev

arcie dev validates the agent, prints any diagnostics, and starts a local HTTP server on port 3000. Send it a message — pass stream: true to receive the event stream as newline-delimited JSON:

curl -N -X POST http://localhost:3000 \
  -H "Content-Type: application/json" \
  -d '{"message": "Hello!", "stream": true}'

You'll see Zett protocol events — session.started, message.appended deltas, message.completed, and so on.

5. Add a tool

Drop a file in agent/tools/. Its default export is the tool — no registration:

// agent/tools/get_weather.ts
import { defineTool } from "arcie/tools";
import { z } from "zod";
 
export default defineTool({
  description: "Get the current weather for a city.",
  inputSchema: z.object({ city: z.string() }),
  async execute({ city }) {
    return { city, condition: "Sunny", temperatureF: 72 };
  },
});

Restart arcie dev and the tool count updates. The file name (get_weather) becomes the tool name. Learn more in Tools.

6. Build for production

npm run build

arcie build compiles the agent into a manifest at .arcie/manifest.json — the artifact Cencori runs. See the CLI reference.

Where to go next