Models

Models are provider-backed capability objects. Each model implements one Anvia contract, allowing the rest of the application to depend on the capability instead of a provider SDK.

Choose the model family

Anvia separates model families because they accept different inputs and return different results:

• Completion models generate assistant content, stream text, call tools, and power agents.
• Embedding models turn text into dense vectors for search and retrieval.
• Image generation models create image bytes from a prompt.
• Audio generation models create speech bytes from text.
• Transcription models convert audio bytes into text.
• OCR models extract text and structure from documents and images.

Provider packages expose only the families supported by their APIs.

Create models at the provider boundary

Keep credentials and provider-specific configuration in server-side application code. A provider client creates the model objects passed to Anvia helpers and agents.

import { OpenAIClient } from '@anvia/openai'

const apiKey = process.env.OPENAI_API_KEY

if (!apiKey) {
  throw new Error('OPENAI_API_KEY is required')
}

const openai = new OpenAIClient({ apiKey })

const completionModel = openai.completionModel({
    modelId: 'gpt-5.6-sol',
    api: "responses"
})
const embeddingModel = openai.embeddingModel({
    modelId: 'text-embedding-3-small'
})
const imageModel = openai.imageGenerationModel({ modelId: 'gpt-image-2' })
const audioModel = openai.speechGenerationModel({ modelId: 'tts-1' })
const transcriptionModel = openai.transcriptionModel({ modelId: 'whisper-1' })

These objects contain the provider implementation. Core helpers receive them explicitly and coordinate the provider-neutral request.

Use the model with a core helper

Each model family has a focused helper. For example, a completion model is passed through the model option:

import { generateCompletion } from '@anvia/core'

const result = await generateCompletion({
    prompt: 'Explain semantic search briefly.',
    model: completionModel,
    instructions: 'Use plain language.'
})

console.log(result.text)

The same dependency-injection pattern applies to embeddings, image generation, speech, and transcription.

Completion capabilities are explicit

Completion models declare the features implemented by that exact adapter:

console.log(completionModel.capabilities)

// {
//   streaming: true,
//   tools: true,
//   toolChoice: true,
//   imageInput: true,
//   documentInput: true,
//   outputSchema: true,
//   reasoning: true,
//   providerTools: true,
// }

The shown values are illustrative. Read the actual object at runtime and smoke test the exact model ID, provider account, region, endpoint, and feature combination used by the product.

Provider support differs

Anvia v1 currently includes these media and data adapters:

• OpenAI: completions, embeddings, image generation, speech, and transcription.
• Gemini: completions, embeddings, image generation, and transcription.
• Grok: completions, image generation, speech, and transcription.
• Mistral: completions, embeddings, and OCR.
• @anvia/transformers: local embeddings with Transformers.js models.

Use the provider capability matrix to narrow the options, then verify the exact workflow before production.

Keep model creation centralized

Create provider clients and models in a configuration or dependency module rather than inside product components. Central construction makes credentials, model selection, endpoint overrides, and tests easier to control.

Continue with Completion models.