OpenRouter Plugin

Multi-provider LLM access through OpenRouter

Features

• Multiple providers - Access 50+ models from various providers
• Automatic failover - Route to available providers
• Cost optimization - Choose models by price/performance
• Single API key - One key for all providers

Installation

elizaos plugins add @elizaos/plugin-openrouter

Configuration

Environment Variables

# Required
OPENROUTER_API_KEY=sk-or-...

# Optional model configuration
# You can use any model available on OpenRouter
OPENROUTER_SMALL_MODEL=google/gemini-2.0-flash-001           # Default: google/gemini-2.0-flash-001
OPENROUTER_LARGE_MODEL=google/gemini-2.5-flash-preview-05-20 # Default: google/gemini-2.5-flash-preview-05-20
OPENROUTER_IMAGE_MODEL=anthropic/claude-3-5-sonnet           # For vision tasks

# Examples of other available models:
# OPENROUTER_SMALL_MODEL=anthropic/claude-3-haiku
# OPENROUTER_LARGE_MODEL=anthropic/claude-3-opus
# OPENROUTER_LARGE_MODEL=openai/gpt-4o
# OPENROUTER_SMALL_MODEL=meta-llama/llama-3.1-8b-instruct:free

Character Configuration

{
  "name": "MyAgent",
  "plugins": [
    "@elizaos/plugin-openrouter",
    "@elizaos/plugin-ollama"  // For embeddings
  ]
}

Supported Operations

Operation	Support	Notes
TEXT_GENERATION	✅	All available models
EMBEDDING	❌	Use fallback plugin
OBJECT_GENERATION	✅	Model dependent

Important: Embedding Fallback

OpenRouter doesn’t provide embedding endpoints, so include a fallback:

{
  "plugins": [
    "@elizaos/plugin-openrouter",  // Text generation
    "@elizaos/plugin-openai"        // Embeddings
  ]
}

Model Configuration

The plugin uses model tiers:

• SMALL_MODEL: Fast, cost-effective responses
• LARGE_MODEL: Complex reasoning, best quality
• IMAGE_MODEL: Multimodal capabilities

OpenRouter provides access to 50+ models from various providers. You can use:

Premium Models

• Any Anthropic Claude model (Opus, Sonnet, Haiku)
• Any OpenAI GPT model (GPT-4o, GPT-4, GPT-3.5)
• Google Gemini models (Pro, Flash, etc.)
• Cohere Command models

Open Models

• Meta Llama models (3.1, 3.2, 3.3)
• Mistral/Mixtral models
• Many models with :free suffix for testing

Pricing Strategy

OpenRouter charges a small markup (usually ~10%) on top of provider prices:

1. Pay-per-token - No monthly fees
2. Price transparency - See costs per model
3. Credits system - Pre-pay for usage

External Resources

• Plugin Source
• OpenRouter Documentation
• Model List & Pricing