Build an agent

Create an agent after a direct completion confirms that the provider model works.

Create the agent

import { AgentBuilder } from '@anvia/core'
import { OpenAIClient } from '@anvia/openai'

const openai = new OpenAIClient({
  apiKey: process.env.OPENAI_API_KEY,
})

const model = openai.completionModel('gpt-5')

export const agent = new AgentBuilder('support', model)
  .name('Support')
  .description('Answers customer support questions.')
  .instructions(
    'Answer clearly and ask for any missing details.',
  )
  .defaultMaxTurns(4)
  .build()

The agent depends on Anvia's completion-model contract, not the OpenAI API. Swap the provider model at the construction boundary without rewriting agent behavior.

Send the first prompt

const response = await agent
  .prompt('What information do you need to investigate a failed checkout?')
  .send()

console.log(response.output)

send() returns the final output, accumulated usage, messages created during the run, and trace metadata when tracing is enabled.

Keep the identity stable

The first AgentBuilder argument is the agent ID. Keep it stable because sessions, traces, event records, evaluations, and development tooling can use it as an identifier.

defaultMaxTurns(4) bounds the model-and-tool loop. A smaller request-level limit can override it for one run.