Dashboard|

Quickstart

Run your first agent with a tool in five minutes.

1. Install

 
npm install cencori zod
# .env
CENCORI_API_KEY=csk_...

Get a key from your project's API keys page. Use a secret key: the SDK runs on your server.

2. Define a tool and an agent

 
// agent.ts
import { Agent, tool } from 'cencori/agents';
import { z } from 'zod';
 
export const getWeather = tool({
    name: 'get_weather',
    description: 'Current weather for a city.',
    parameters: z.object({ city: z.string() }),
    execute: async ({ city }) => ({ city, tempC: 31, sky: 'sunny' }),
});
 
export const assistant = new Agent({
    name: 'Assistant',
    instructions: 'Answer briefly. Use tools for live data.',
    tools: [getWeather],
});

No model means the default, GPT OSS 120B on Cerebras: fast and good at calling tools. Set model to any model in the catalog.

3. Run it

 
import { run } from 'cencori/agents';
import { assistant } from './agent';
 
const result = await run(assistant, "What's the weather in Lagos?");
console.log(result.finalOutput); // "It's 31°C and sunny in Lagos."

The model asked for get_weather, the SDK ran it in your process, sent the result back, and the model answered.

4. Stream it

 
const stream = run(assistant, "What's the weather in Lagos?", { stream: true });
for await (const event of stream) {
    if (event.type === 'text_delta') process.stdout.write(event.delta);
}
const result = await stream.completed;

5. Keep the conversation going

 
const first = await run(assistant, "What's the weather in Lagos?");
const second = await run(assistant, 'And in Nairobi?', { sessionId: first.sessionId });

6. See the trace

Open Observability → Traces in your project. Every run is there with its turns, tool calls, timings and cost.

Next
Did you like the content?