Getting Started with OpenRouter

Introduction

TPipe-OpenRouter provides access to OpenRouter’s unified API that aggregates 300+ LLM models from various providers. It offers a single, consistent interface to models from Anthropic, OpenAI, Google, Meta, DeepSeek, Mistral, and many others.

Prerequisites

1. OpenRouter API Key: Obtain an API key from openrouter.ai. You can get started with free models without adding credits.
2. Project Dependency: Add the :TPipe-OpenRouter module to your Gradle configuration.

Basic Usage

To get started, create an OpenRouterPipe instance and configure your model and API key.

import openrouterPipe.OpenRouterPipe
import kotlinx.coroutines.runBlocking

fun main() = runBlocking {
    val pipe = OpenRouterPipe()
        .setModel("anthropic/claude-3.5-sonnet")
        .setApiKey("sk-or-...")
        .setTemperature(0.7)
        .setSystemPrompt("You are a helpful assistant.")

    pipe.init()

    val result = pipe.execute("What is the capital of France?")
    println(result.text)
}

Using Free Models

OpenRouter provides free access to certain models via the special openrouter/free routing. This allows experimentation without incurring costs.

val pipe = OpenRouterPipe()
    .setModel("openrouter/free")
    .setApiKey("sk-or-...")  // Still required for identification
    .setHttpReferer("https://yourapp.com")
    .setOpenRouterTitle("My Application")

Note: Even with free models, the API key is required for request identification and tracking.

Advanced Configuration

Provider Preferences

Request routing through specific providers:

val pipe = OpenRouterPipe()
    .setModel("openai/gpt-4o")
    .setProviderPreferences(
        env.ProviderPreferences(
            order = listOf("openai", "anthropic"),
            allow_fallbacks = true
        )
    )

Reasoning Models

For models that support reasoning (like Claude with extended thinking):

val pipe = OpenRouterPipe()
    .setModel("anthropic/claude-3.5-sonnet")
    .setReasoningEffort("high")
    // or with full config:
    .setReasoningConfig(
        env.ReasoningConfig(
            effort = "high",
            maxTokens = 8192
        )
    )

Caching and Service Tiers

Enable request caching for improved performance:

val pipe = OpenRouterPipe()
    .setModel("anthropic/claude-3.5-sonnet")
    .setCacheControl("5m")  // Cache for 5 minutes
    .setServiceTier("priority")  // Priority request handling

Structured Outputs

Request JSON schema-validated responses:

val pipe = OpenRouterPipe()
    .setModel("openai/gpt-4o")
    .setResponseFormat("json_schema", jsonObjectOf(
        "type" to "object",
        "properties" to jsonObjectOf(
            "name" to jsonObjectOf("type" to "string"),
            "age" to jsonObjectOf("type" to "number")
        ),
        "required" to listOf("name", "age")
    ))
    .setStructuredOutputs(true)

Comparison with Other Providers

Feature	BedrockPipe	OllamaPipe	OpenRouterPipe
Provider	AWS Bedrock	Local Ollama	OpenRouter (Multi-provider)
Model Access	AWS-hosted	Local models	300+ models
Reasoning	Native support	<think> extraction	Extended thinking
Tool Calling	Converse API	Native tools	OpenAI-compatible
Streaming	Stream handler	Ktor async	Server-Sent Events
Free Tier	No	Yes (local)	Yes (limited models)

Next Steps

• OpenRouterPipe Class API - Full API reference for OpenRouterPipe.
• Pipe Class API - Core pipe abstraction.