Model types and configuration

For exact v4 signatures and all public model exports, use the models and messages API reference. This page explains selected types. Factory examples configure adapters; they do not call providers. Install each adapter’s optional dependency from its guide. Current text-model choices are maintained in example models.

ModelProvider Interface

Every model in Agentium implements this interface. Use a built-in factory such as openai(), anthropic(), or google(), or implement a custom provider. This page explains common contracts; the installed v4 declarations define the full option surface.

Property/Method	Type	Description
`providerId`	`string` (readonly)	Provider identifier (e.g. `"openai"`, `"anthropic"`, `"google"`)
`modelId`	`string` (readonly)	Model identifier (e.g. `"gpt-6.1-sol"`, `"claude-sonnet-5-5"`)
`generate(messages, options?)`	`Promise<ModelResponse>`	Send messages, get a complete response
`stream(messages, options?)`	`AsyncGenerator<StreamChunk>`	Send messages, get a streaming response

ModelConfig

Options passed to generate() and stream(). Only options exposed by RunOpts can be passed directly to agent.run(). A ModelConfig field is not automatically a supported run option.

Property	Type	Default	Description
`temperature`	`number`	Provider default	Sampling control; supported ranges vary, and lower temperature does not guarantee deterministic output
`maxTokens`	`number`	Provider default	Maximum tokens in the response
`topP`	`number`	`undefined`	Nucleus sampling. Alternative to temperature
`stop`	`string[]`	`undefined`	Stop sequences — model stops generating when it produces any of these
`responseFormat`	`"text" \| "json" \| { type: "json_schema"; schema: object; name?: string }`	`"text"`	`"text"` = plain text, `"json"` = JSON mode, or a JSON schema for structured output
`apiKey`	`string`	Provider-level key	Per-request API key override
`reasoning`	`ReasoningConfig`	`undefined`	Extended thinking. OpenAI also takes `summary`, `mode`, `context`
`providerOptions`	`ModelConfig["providerOptions"]`	`undefined`	`promptCache`, `promptCacheRetention`, `compactionTokens`, `clearToolResults`, `mediaResolution`, `cachedContent`, `googleSearch`
`signal`	`AbortSignal`	`undefined`	Cooperative cancellation forwarded by compatible providers
`questions`	`Record<string, unknown>`	`undefined`	Jev questions for this call; ignored by chat providers

Factory Functions and Provider Configs

Provider factories create a ModelProvider. The tables below describe commonly used options; individual provider guides cover integration-specific behavior. Model identifiers are examples, not an availability guarantee.

openai(modelId, config?)

import { openai } from "@agentium/core";
const model = openai(process.env.OPENAI_MODEL ?? "gpt-6.1-sol", { apiKey: "sk-..." });

Config Property	Type	Default	Description
`apiKey`	`string`	`OPENAI_API_KEY` env var	OpenAI API key
`baseURL`	`string`	`"https://api.openai.com/v1"`	Custom base URL. Use this for OpenAI-compatible APIs (Together, Groq, etc.)

anthropic(modelId, config?)

import { anthropic } from "@agentium/core";
const model = anthropic(process.env.ANTHROPIC_MODEL ?? "claude-sonnet-5-5", { apiKey: "sk-ant-..." });

Config Property	Type	Default	Description
`apiKey`	`string`	`ANTHROPIC_API_KEY` env var	Anthropic API key

google(modelId, config?)

import { google } from "@agentium/core";
const model = google(process.env.GOOGLE_MODEL ?? "gemini-3.8-flash", { apiKey: "AI..." });

Config Property	Type	Default	Description
`apiKey`	`string`	`GOOGLE_API_KEY` env var	Google AI Studio API key

vertex(modelId, config?)

import { vertex } from "@agentium/core";
const model = vertex("gemini-2.5-flash", {
  project: "my-gcp-project",
  location: "us-central1",
});

Config Property	Type	Default	Description
`project`	`string`	`GOOGLE_CLOUD_PROJECT` env var	GCP project ID
`location`	`string`	`"us-central1"`	GCP region
`credentials`	`string`	Application Default Credentials	Path to a service account JSON key file, OR the JSON string itself

ollama(modelId, config?)

import { ollama } from "@agentium/core";
const model = ollama("llama3.2", { host: "http://localhost:11434" });

Config Property	Type	Default	Description
`host`	`string`	`"http://localhost:11434"`	Ollama server URL

deepseek(modelId, config?)

import { deepseek } from "@agentium/core";
const model = deepseek("deepseek-chat");

Config Property	Type	Default	Description
`apiKey`	`string`	`DEEPSEEK_API_KEY` env var	DeepSeek API key

mistral(modelId, config?)

import { mistral } from "@agentium/core";
const model = mistral("mistral-large-latest");

Config Property	Type	Default	Description
`apiKey`	`string`	`MISTRAL_API_KEY` env var	Mistral API key

xai(modelId, config?)

import { xai } from "@agentium/core";
const model = xai("grok-3");

Config Property	Type	Default	Description
`apiKey`	`string`	`XAI_API_KEY` env var	xAI API key

perplexity(modelId, config?)

import { perplexity } from "@agentium/core";
const model = perplexity("sonar-pro");

Config Property	Type	Default	Description
`apiKey`	`string`	`PERPLEXITY_API_KEY` env var	Perplexity API key

jev(modelId?, config?)

Not a chat model. Pass questions on agent.run(input, { questions }), or fall back to constructor questions / structuredOutput / closed-set tools. See Jev.

import { Agent, jev, choice, noul, score } from "@agentium/core";

const agent = new Agent({ name: "triage", model: jev("jev-latest") });
const ticket = "My paid invoice still appears unpaid.";
try {
const result = await agent.run(ticket, {
  questions: {
    urgent: noul("Is this urgent?"),
    team: choice("Which team?", { billing: null, tech: null }),
    severity: score("Impact?", ["low", "high"]),
  },
});
console.log(result.structured);
} finally {
  await agent.close();
}

Config Property	Type	Default	Description
`apiKey`	`string`	`TYPESAFE_API_KEY` env var	TypeSafe API key
`baseURL`	`string`	`TYPESAFE_BASE_URL` / `https://api.typesafe.ai`	API root
`questions`	`Record<string, Question>`	—	Default `choice` / `noul` / `score` if a run does not pass its own. Per-run `questions` replace these.

Helpers (also exported from @agentium/core): choice(instructions, { label: desc \| null }), noul(instructions?, { true?, false? }), score(instructions, levels[]) — score is 0-indexed, lowest first.

cohere(modelId, config?)

import { cohere } from "@agentium/core";
const model = cohere("command-r-plus");

Config Property	Type	Default	Description
`apiKey`	`string`	`COHERE_API_KEY` env var	Cohere API key

meta(modelId, config?)

import { meta } from "@agentium/core";
const model = meta("Llama-4-Maverick-17B-128E-Instruct-FP8");

Config Property	Type	Default	Description
`apiKey`	`string`	`META_API_KEY` env var	Meta Llama API key

awsBedrock(modelId, config?)

import { awsBedrock } from "@agentium/core";
const model = awsBedrock("amazon.nova-pro-v1:0", {
  region: "us-east-1",
});

Config Property	Type	Default	Description
`region`	`string`	`AWS_REGION` env var	AWS region
`accessKeyId`	`string`	AWS credential chain	AWS access key
`secretAccessKey`	`string`	AWS credential chain	AWS secret key
`sessionToken`	`string`	`undefined`	AWS session token (for temporary credentials)

awsClaude(modelId, config?)

import { awsClaude } from "@agentium/core";
const model = awsClaude(process.env.AWS_CLAUDE_MODEL ?? "anthropic.claude-sonnet-4-20250514-v1:0", {
  awsRegion: "us-east-1",
});

Config Property	Type	Default	Description
`awsRegion`	`string`	`AWS_REGION` env var	AWS region
`awsAccessKey`	`string`	AWS credential chain	AWS access key
`awsSecretKey`	`string`	AWS credential chain	AWS secret key
`awsSessionToken`	`string`	`AWS_SESSION_TOKEN` env var	Temporary credential session token

azureOpenai(modelId, config?)

import { azureOpenai } from "@agentium/core";
const model = azureOpenai("gpt-4o", {
  endpoint: "https://my-resource.openai.azure.com",
  deployment: "my-deployment",
  apiKey: "...",
});

Config Property	Type	Default	Description
`endpoint`	`string`	`AZURE_OPENAI_ENDPOINT`	Full Azure OpenAI endpoint URL
`deployment`	`string`	`AZURE_OPENAI_DEPLOYMENT`	Your configured deployment name
`apiKey`	`string`	`AZURE_OPENAI_API_KEY` env var	Azure API key
`apiVersion`	`string`	`AZURE_OPENAI_API_VERSION` or `"2024-10-21"`	Azure API version

azureFoundry(modelId, config?)

import { azureFoundry } from "@agentium/core";
const model = azureFoundry("Phi-4", {
  endpoint: "https://my-endpoint.inference.ai.azure.com",
  apiKey: "...",
});

Config Property	Type	Default	Description
`endpoint`	`string`	Required	Azure AI Foundry inference endpoint
`apiKey`	`string`	`AZURE_FOUNDRY_API_KEY` env var	Azure API key

vercel(modelId, config?)

import { vercel } from "@agentium/core";
const model = vercel("v0-1.0-md");

Config Property	Type	Default	Description
`apiKey`	`string`	`VERCEL_API_KEY` env var	Vercel API key

OpenAI-Compatible Providers

Any API that follows the OpenAI API format can be used with the openai() factory by setting baseURL:

import { openai } from "@agentium/core";

// Together AI
const together = openai("meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo", {
  apiKey: process.env.TOGETHER_API_KEY,
  baseURL: "https://api.together.xyz/v1",
});

// Groq
const groq = openai("llama-3.3-70b-versatile", {
  apiKey: process.env.GROQ_API_KEY,
  baseURL: "https://api.groq.com/openai/v1",
});

// Fireworks AI
const fireworks = openai("accounts/fireworks/models/llama-v3p1-70b-instruct", {
  apiKey: process.env.FIREWORKS_API_KEY,
  baseURL: "https://api.fireworks.ai/inference/v1",
});

// OpenRouter
const openrouter = openai("anthropic/claude-sonnet-4-20250514", {
  apiKey: process.env.OPENROUTER_API_KEY,
  baseURL: "https://openrouter.ai/api/v1",
});

// LM Studio (local)
const lmstudio = openai("local-model", {
  baseURL: "http://localhost:1234/v1",
});

Environment Variables

Quick reference for all provider environment variables:

Provider	Env Variable	Factory
OpenAI	`OPENAI_API_KEY`	`openai()`
Anthropic	`ANTHROPIC_API_KEY`	`anthropic()`
Google	`GOOGLE_API_KEY`	`google()`
Vertex AI	`GOOGLE_CLOUD_PROJECT`	`vertex()`
Ollama	— (no key needed)	`ollama()`
DeepSeek	`DEEPSEEK_API_KEY`	`deepseek()`
Mistral	`MISTRAL_API_KEY`	`mistral()`
xAI	`XAI_API_KEY`	`xai()`
Perplexity	`PERPLEXITY_API_KEY`	`perplexity()`
Jev (TypeSafe)	`TYPESAFE_API_KEY`	`jev()`
Cohere	`COHERE_API_KEY`	`cohere()`
Meta	`META_API_KEY`	`meta()`
AWS	`AWS_REGION`, `AWS_ACCESS_KEY_ID`, `AWS_SECRET_ACCESS_KEY`	`awsBedrock()`, `awsClaude()`
Azure OpenAI	`AZURE_OPENAI_API_KEY`	`azureOpenai()`
Azure Foundry	`AZURE_FOUNDRY_API_KEY`	`azureFoundry()`
Vercel	`VERCEL_API_KEY`	`vercel()`

Reference

Package APIs

Type guides

Upgrades

Architecture and tooling

Model types and configuration

ModelProvider Interface

ModelConfig

Factory Functions and Provider Configs

openai(modelId, config?)

anthropic(modelId, config?)

google(modelId, config?)

vertex(modelId, config?)

ollama(modelId, config?)

deepseek(modelId, config?)

mistral(modelId, config?)

xai(modelId, config?)

perplexity(modelId, config?)

jev(modelId?, config?)

cohere(modelId, config?)

meta(modelId, config?)

awsBedrock(modelId, config?)

awsClaude(modelId, config?)

azureOpenai(modelId, config?)

azureFoundry(modelId, config?)

vercel(modelId, config?)

OpenAI-Compatible Providers

Environment Variables

​ModelProvider Interface

​ModelConfig

​Factory Functions and Provider Configs

​openai(modelId, config?)

​anthropic(modelId, config?)

​google(modelId, config?)

​vertex(modelId, config?)

​ollama(modelId, config?)

​deepseek(modelId, config?)

​mistral(modelId, config?)

​xai(modelId, config?)

​perplexity(modelId, config?)

​jev(modelId?, config?)

​cohere(modelId, config?)

​meta(modelId, config?)

​awsBedrock(modelId, config?)

​awsClaude(modelId, config?)

​azureOpenai(modelId, config?)

​azureFoundry(modelId, config?)

​vercel(modelId, config?)

​OpenAI-Compatible Providers

​Environment Variables

ModelProvider Interface

ModelConfig

Factory Functions and Provider Configs

openai(modelId, config?)

anthropic(modelId, config?)

google(modelId, config?)

vertex(modelId, config?)

ollama(modelId, config?)

deepseek(modelId, config?)

mistral(modelId, config?)

xai(modelId, config?)

perplexity(modelId, config?)

jev(modelId?, config?)

cohere(modelId, config?)

meta(modelId, config?)

awsBedrock(modelId, config?)

awsClaude(modelId, config?)

azureOpenai(modelId, config?)

azureFoundry(modelId, config?)

vercel(modelId, config?)

OpenAI-Compatible Providers

Environment Variables