Using Different Providers

Connect to LLM providers other than OpenRouter—OpenAI direct, Anthropic direct, or self-hosted endpoints. This guide covers custom base URLs and provider-specific configuration.

Prerequisites

Before starting, you should:

Know the basics of creating a client
Have API credentials for your target provider

Overview

We'll use different providers by:

Setting a custom baseUrl
Using provider-specific model names
Adjusting headers if needed

Step 1: Set the Base URL

Pass baseUrl when creating the client to point to a different API:

import { createLLMClient } from '@motioneffector/llm'

const client = createLLMClient({
  apiKey: process.env.OPENAI_KEY!,
  model: 'gpt-4o',
  baseUrl: 'https://api.openai.com/v1'
})

The client appends /chat/completions to the base URL, so provide just the base path.

Step 2: Use Provider Model Names

Each provider has its own model naming:

// OpenAI
const openai = createLLMClient({
  apiKey: process.env.OPENAI_KEY!,
  model: 'gpt-4o',  // Not 'openai/gpt-4o'
  baseUrl: 'https://api.openai.com/v1'
})

// Anthropic
const anthropic = createLLMClient({
  apiKey: process.env.ANTHROPIC_KEY!,
  model: 'claude-sonnet-4-20250514',  // Anthropic's format
  baseUrl: 'https://api.anthropic.com/v1'
})

Step 3: Handle Provider Differences

OpenRouter headers (HTTP-Referer, X-Title) are only sent when the URL contains "openrouter". Other providers don't need them.

The library handles standard OpenAI-compatible APIs. Providers with different formats may need additional configuration or may not be compatible.

Complete Example

import { createLLMClient } from '@motioneffector/llm'

// OpenAI Direct
const openai = createLLMClient({
  apiKey: process.env.OPENAI_KEY!,
  model: 'gpt-4o',
  baseUrl: 'https://api.openai.com/v1'
})

const response = await openai.chat([
  { role: 'user', content: 'Hello from OpenAI!' }
])

console.log(response.content)

Variations

OpenAI

const client = createLLMClient({
  apiKey: process.env.OPENAI_KEY!,
  model: 'gpt-4o',
  baseUrl: 'https://api.openai.com/v1'
})

Available models: gpt-4o, gpt-4-turbo, gpt-3.5-turbo, etc.

Azure OpenAI

Azure uses a different URL format with deployments:

const client = createLLMClient({
  apiKey: process.env.AZURE_OPENAI_KEY!,
  model: 'gpt-4',  // Deployment name
  baseUrl: 'https://YOUR-RESOURCE.openai.azure.com/openai/deployments/YOUR-DEPLOYMENT'
})

Note: Azure may require additional headers. Check Azure documentation for current requirements.

Local Models (Ollama, LM Studio)

Connect to locally-running models:

// Ollama
const ollama = createLLMClient({
  apiKey: 'ollama',  // Ollama doesn't require a key
  model: 'llama2',
  baseUrl: 'http://localhost:11434/v1'
})

// LM Studio
const lmstudio = createLLMClient({
  apiKey: 'lm-studio',
  model: 'local-model',
  baseUrl: 'http://localhost:1234/v1'
})

Together.ai

const client = createLLMClient({
  apiKey: process.env.TOGETHER_KEY!,
  model: 'meta-llama/Llama-3-70b-chat-hf',
  baseUrl: 'https://api.together.xyz/v1'
})

Groq

const client = createLLMClient({
  apiKey: process.env.GROQ_KEY!,
  model: 'llama3-70b-8192',
  baseUrl: 'https://api.groq.com/openai/v1'
})

Multiple Providers in One App

Create separate clients for different providers:

const openai = createLLMClient({
  apiKey: process.env.OPENAI_KEY!,
  model: 'gpt-4o',
  baseUrl: 'https://api.openai.com/v1'
})

const anthropic = createLLMClient({
  apiKey: process.env.ANTHROPIC_KEY!,
  model: 'claude-sonnet-4-20250514',
  baseUrl: 'https://api.anthropic.com/v1'
})

// Use the appropriate client
const response = await openai.chat(messages)

Troubleshooting

404 Not Found

Symptom: Request returns 404.

Cause: Wrong base URL or model name.

Solution: Check the provider's documentation for the correct endpoint. The library appends /chat/completions, so don't include that in the base URL.

Authentication Failed

Symptom: 401 or 403 error.

Cause: Wrong API key or key format.

Solution: Verify the key is for the correct provider. Some providers use different key formats (Bearer token vs API key header).

Unsupported Response Format

Symptom: Parse errors or missing content.

Cause: Provider uses non-standard response format.

Solution: The library expects OpenAI-compatible responses. Providers with different formats may not work without modification.

Headers Not Sent

Symptom: Provider rejects requests for missing headers.

Cause: Custom headers needed for this provider.

Solution: Currently the library doesn't support custom headers. Consider using OpenRouter as a proxy instead, which handles provider-specific requirements.