The langchain package provides a universal interface for initializing chat models from any supported provider. This allows you to easily switch between different model providers without changing your code structure.
Core Functions
initChatModel
Initializes a chat model from a model name and optional provider. The function automatically infers the provider when possible and handles dynamic model loading.
import { initChatModel } from "langchain";
const model = await initChatModel("openai:gpt-4o", {
temperature: 0.7,
});
const result = await model.invoke("Hello, how are you?");
The model name, optionally prefixed with provider (e.g., "openai:gpt-4o", "anthropic:claude-3-opus-20240229", or just "gpt-4o").
Explicit model provider. Supported values:
openai - OpenAI models via @langchain/openaianthropic - Anthropic models via @langchain/anthropicazure_openai - Azure OpenAI via @langchain/openaigoogle-vertexai - Google Vertex AI via @langchain/google-vertexaigoogle-vertexai-web - Google Vertex AI Web via @langchain/google-vertexai-webgoogle-genai - Google Generative AI via @langchain/google-genaibedrock - AWS Bedrock via @langchain/awscohere - Cohere via @langchain/coheremistralai - Mistral AI via @langchain/mistralaigroq - Groq via @langchain/groqollama - Ollama via @langchain/ollamacerebras - Cerebras via @langchain/cerebrasdeepseek - DeepSeek via @langchain/deepseekxai - xAI via @langchain/xaifireworks - Fireworks via @langchain/community/chat_models/fireworkstogether - Together AI via @langchain/community/chat_models/togetheraiperplexity - Perplexity via @langchain/community/chat_models/perplexity
fields.configurableFields
Which model parameters are configurable at runtime:
undefined - No configurable fields (default)"any" - All fields are configurable (⚠️ Security Note: allows changing apiKey, baseUrl, etc.)string[] - Specific fields that can be configured
Prefix for configurable fields. Useful for namespacing configuration when using multiple models.
Override profiling information for the model (e.g., token limits, capabilities). If not provided, the profile is inferred from the model instance.
Additional parameters passed to the underlying chat model constructor (e.g., temperature, maxTokens, apiKey).
Promise<ConfigurableModel> - A configurable chat model instance.
Source: libs/langchain/src/chat_models/universal.ts:719
Examples
Basic Usage
import { initChatModel } from "langchain";
// OpenAI
const gpt4 = await initChatModel("openai:gpt-4o", {
temperature: 0.25,
});
const gpt4Result = await gpt4.invoke("What's your name?");
// Anthropic
const claude = await initChatModel("anthropic:claude-3-opus-20240229", {
temperature: 0.25,
});
const claudeResult = await claude.invoke("What's your name?");
// Google Vertex AI (with explicit provider)
const gemini = await initChatModel("gemini-1.5-pro", {
modelProvider: "google-vertexai",
temperature: 0.25,
});
const geminiResult = await gemini.invoke("What's your name?");
Configurable Models
Create models that can be reconfigured at runtime:
import { initChatModel } from "langchain";
// Partially configurable model
const configurableModel = await initChatModel(undefined, {
temperature: 0,
configurableFields: ["model", "apiKey"],
});
// Use with GPT-4
const gpt4Result = await configurableModel.invoke("What's your name?", {
configurable: {
model: "gpt-4",
},
});
// Use with Claude
const claudeResult = await configurableModel.invoke("What's your name?", {
configurable: {
model: "claude-sonnet-4-5-20250929",
},
});
Fully Configurable with Prefix
import { initChatModel } from "langchain";
const configurableModel = await initChatModel("gpt-4", {
modelProvider: "openai",
configurableFields: "any",
configPrefix: "llm",
temperature: 0,
});
// Use default (GPT-4)
const openaiResult = await configurableModel.invoke("What's your name?", {
configurable: {
llm_apiKey: process.env.OPENAI_API_KEY,
},
});
// Switch to Claude at runtime
const claudeResult = await configurableModel.invoke("What's your name?", {
configurable: {
llm_model: "claude-sonnet-4-5-20250929",
llm_modelProvider: "anthropic",
llm_temperature: 0.6,
llm_apiKey: process.env.ANTHROPIC_API_KEY,
},
});
import { initChatModel } from "langchain";
import { z } from "zod";
import { tool } from "@langchain/core/tools";
const getWeatherTool = tool(
(input) => JSON.stringify(input),
{
schema: z.object({
location: z.string().describe("The city and state, e.g. San Francisco, CA"),
}).describe("Get the current weather in a given location"),
name: "GetWeather",
description: "Get the current weather in a given location",
}
);
const model = await initChatModel("gpt-4", {
configurableFields: ["model", "modelProvider", "apiKey"],
temperature: 0,
});
const modelWithTools = model.bindTools([getWeatherTool]);
const result = await modelWithTools.invoke(
"What's the weather in San Francisco?",
{
configurable: {
apiKey: process.env.OPENAI_API_KEY,
},
}
);
Custom Model Profile
import { initChatModel } from "langchain";
const model = await initChatModel("gpt-4o-mini", {
profile: {
maxInputTokens: 100000,
maxOutputTokens: 4096,
supportsToolCalling: true,
supportsStructuredOutput: true,
},
});
console.log(model.profile);
Provider Inference
The function automatically infers the provider from model name prefixes:
gpt-3..., gpt-4..., gpt-5..., o1..., o3..., o4... → openaiclaude... → anthropiccommand... → cohereaccounts/fireworks... → fireworksgemini... → google-vertexaiamazon.... → bedrockmistral... → mistralaisonar..., pplx... → perplexity
ConfigurableModel
The ConfigurableModel class returned by initChatModel extends BaseChatModel and provides:
Methods
invoke(input, options?) - Invoke the model with a single inputstream(input, options?) - Stream the model’s responsebatch(inputs, options?) - Process multiple inputs in batchbindTools(tools, params?) - Bind tools to the modelwithStructuredOutput(schema) - Configure structured outputwithConfig(config) - Bind configuration to the model
Properties
profile - Model profiling information (token limits, capabilities)
Security Considerations
Setting configurableFields: "any" allows runtime modification of sensitive fields like apiKey and baseUrl. This could redirect model requests to a different service or user.Always enumerate specific configurableFields when accepting untrusted configurations.
Type Exports
ConfigurableModel
The model class returned by initChatModel. Extends BaseChatModel with configuration support.
ConfigurableChatModelCallOptions
Call options for configurable models, including tool binding support.
ChatModelProvider
Union type of all supported model providers.
- Agents - Using models with agents
- Tools - Creating tools for chat models
- Messages - Working with message types
