Create a completion

Call a completion model directly when the application owns everything before and after the model request.

Before you start

Complete Install and setup and create a completion model.

Send a request

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

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

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

const result = await createCompletion(model, {
  instructions: 'Answer clearly and concisely.',
  input: 'Summarize Anvia in one sentence.',
  maxTokens: 120,
})

console.log(result.text)

Configure the request

Option	Purpose
input	A string, one message, or several messages appended to the transcript.
messages	Existing provider-neutral message history.
instructions	Stable behavior for this request.
documents	Small text documents that should accompany the request.
temperature	Provider-supported output randomness.
maxTokens	Maximum generated tokens.
params	Provider-specific options.

At least one of input or messages is required.

Continue an existing transcript

const result = await createCompletion(model, {
  messages: previousMessages,
  input: 'Rewrite the last answer as three bullets.',
})

This sends the supplied transcript but does not persist it. Your application owns the history; use memory and an agent session when Anvia should load and append messages automatically.

Handle failures

The call rejects for empty input, unsupported model features, authentication failures, transport errors, or provider validation errors. Map those errors at the application boundary instead of returning raw provider details to users.