Getting started

This quickstart verifies a provider model, wraps it in an agent, and runs one prompt. You need an ESM-compatible TypeScript project, pnpm, and an OpenAI API key.

Install the runtime

pnpm add @anvia/core @anvia/openai

Set your provider credentials in the environment:

export OPENAI_API_KEY=...

Verify the model

Call the model directly before adding agent behavior. This isolates credentials and provider setup from the rest of the runtime.

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

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

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

const result = await createCompletion(model, {
  instructions: 'Answer clearly and concisely.',
  input: 'Summarize Anvia in one sentence.',
})

console.log(result.text)

createCompletion returns visible text, normalized content, token usage, and the full normalized response. It does not run tools, save memory, or loop through agent turns.

Build an agent

Once the provider works, wrap the model in reusable runtime behavior:

import { AgentBuilder } from '@anvia/core'

const agent = new AgentBuilder('support', model)
  .instructions('Answer support questions clearly and ask for missing details.')
  .defaultMaxTurns(4)
  .build()

The agent depends on the provider-neutral model interface. Changing providers does not require moving provider-specific code into the agent.

Send a prompt

const response = await agent
  .prompt('Explain what the Anvia runtime owns.')
  .send()

console.log(response.output)

The response includes the final output, accumulated usage, run messages, and trace metadata when tracing is enabled.

Stream a response

Use the same prompt with stream() when a UI or CLI should update while the run is active:

for await (const event of agent.prompt('Draft a short launch note.').stream()) {
  if (event.type === 'text_delta') process.stdout.write(event.delta)
  if (event.type === 'final') console.log(event.usage)
}

Agent streams include text, reasoning, tool calls, tool results, turn boundaries, final run metadata, and errors.

Next

Continue with Core concepts, then Build applications when you are ready to expose the agent from a server.