Agents

Agents are the core building blocks of SpinAI. They combine your chosen LLM with actions to solve tasks the way a human decision maker would.

They control what actions to call, what parameters to pass into those actions, all while maintaining state between them and keeping track of the execution history.

​

Creating an agent

To create an agent, use the createAgent function with any of the models supported by Vercel’s AI SDK. Note the model will automatically read your local .env file to authenticate itself.

import { createAgent } from "spinai";
import { openai } from "@ai-sdk/openai";

const calculatorAgent = createAgent({
  instructions: `You are a calculator agent.`,
  actions: [sum, minus],
  model: openai("gpt-4o"),
});

​

Agent Creation Parameters

​

Calling your agent

To call an agent, it’s as simple as using it as a function while passing an object in.

const { response } = await calculatorAgent({
  input: "What is 5 plus 3 - 1?",
});

​

Calling your agent with a custom response format

You can also call your agent with a custom response format by passing in a Zod object.

import { z } from "zod";

const responseSchema = z.object({
  finalNumber: z.number(),
});

const { response } = await calculatorAgent({
  input: "What is 5 plus 3 - 1?",
  responseFormat: responseSchema,
});

console.log(response.finalNumber);

​

Agent Run Parameters

Note: Some parameters are the same as the agent creation parameters, but they can also be passed when calling the agent. Any duplicate parameters passed in to the agent run call will override the ones passed in to the agent creation. This allows you to be flexible with things like using different LLM models, and actions, for an existing agent at different times.

​

Agent Response

The agent response object contains the following properties:

​

Next Steps