Skip to content

How to Use OpenRouter with Simonw's LLM CLI

access dozens of LLM models from a single terminal, including free tier options.

e How to Use OpenRouter with LLM CLI

What is OpenRouter#

  • OpenRouter is a unified API gateway that provides access to dozens of LLM models from different providers through a single endpoint
    • Instead of managing separate API keys for OpenAI, Anthropic, Google, and others, you use one key and one API format
    • OpenRouter handles routing, billing, and rate limiting across providers
    • Includes free tier models (e.g., Google Gemma 4) for testing and experimentation
![](how_to.Use_OpenRouter.md.figs/how_to.Use_OpenRouter.1.png) *Figure 1: OpenRouter acts as a unified API gateway: you manage one API key and one integration point, and OpenRouter handles routing, billing, and rate limiting across all supported LLM providers.*

What is LLM CLI#

  • LLM CLI (by Simon Willison) is a command-line tool for interacting with LLMs:
    • Supports plugins for different backends (OpenAI, Anthropic, OpenRouter, etc.)
    • Provides a consistent interface regardless of the underlying model
    • Version 0.30 at the time of writing
    • Project page: https://github.com/simonw/llm

Installation#

  • Install the llm-openrouter plugin using the LLM CLI plugin system:

    > llm install llm-openrouter

  • Verify the installation:

    > pip show llm-openrouter
Name: llm-openrouter
Version: 0.5
Summary: LLM plugin for models hosted by OpenRouter
Home-page: https://github.com/simonw/llm-openrouter

  • The plugin depends on httpx, llm, and openai Python packages

API Key Configuration#

  • Set your OpenRouter API key as an environment variable:

    > export OPENROUTER_KEY=$OPENROUTER_API_KEY
    • The plugin reads the OPENROUTER_KEY environment variable
    • You can obtain an API key from openrouter.ai/keys
    • Some workflows also use OPENROUTER_API_KEY directly (e.g., for curl requests)

Listing Available Models#

  • List all models available through OpenRouter:

    > llm openrouter models

  • Each model entry includes useful metadata. For example:

    - id: anthropic/claude-opus-4.7
  name: Anthropic: Claude Opus 4.7
  context_length: 1,000,000
  architecture:
    modality: text+image->text
  supports_schema: True

  • After installation, models appear in the main llm models list with an OpenRouter: prefix:

    > llm models | grep openrouter | head

    OpenRouter: openrouter/anthropic/claude-opus-4.7
OpenRouter: openrouter/google/gemma-4-26b-a4b-it:free
OpenRouter: openrouter/google/gemma-4-31b-it:free
OpenRouter: openrouter/qwen/qwen3.6-plus

  • Free models are marked with a :free suffix, making them easy to identify for cost-effective experimentation

Making Requests#

Basic Usage#

  • Send a prompt to any model using the standard llm -m syntax:

    > llm -m openrouter/google/gemma-4-26b-a4b-it:free "Hello!"

    Hello! How can I help you today?

Provider and Routing Options#

  • OpenRouter supports provider-level options, such as sorting by throughput for faster responses:

    > llm -m openrouter/anthropic/claude-opus-4.7 \
    -o provider '{"sort": "throughput"}' \
    "Explain recursion in 1000 words" | tee output.txt
    • The -o provider flag passes a JSON object that configures provider routing
    • {"sort": "throughput"} selects the fastest provider for the requested model
    • tee output.txt saves the response to a file while displaying it in the terminal
![](how_to.Use_OpenRouter.md.figs/how_to.Use_OpenRouter.2.png) *Figure 2: When you pass `-o provider '{"sort": "throughput"}'`, OpenRouter ranks available providers hosting the requested model by response speed and routes to the fastest one, improving latency.*

Using the API Directly with curl#

  • For programmatic access or scripting, you can call the OpenRouter API directly:

    > curl -s https://openrouter.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openrouter/nitro",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'
    • The endpoint is https://openrouter.ai/api/v1/chat/completions (OpenAI-compatible)
    • Use Authorization: Bearer $OPENROUTER_API_KEY for authentication
    • The request body follows the same format as the OpenAI chat completions API

Monitoring Usage#

  • Check your OpenRouter API usage and remaining credits:

    > curl -s https://openrouter.ai/api/v1/auth/key \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  | jq '.data.usage'
    • Returns JSON with usage statistics including credits used and remaining balance
    • Useful for tracking costs across different models and providers

Key Benefits#

  • Single API key: Access dozens of models from different providers with one credential
  • Consistent interface: LLM CLI provides the same commands regardless of backend model
  • Free tier available: Test models like Google Gemma 4 at no cost before committing
  • Provider routing: Route requests to the fastest available provider
  • OpenAI-compatible API: Direct API calls follow the familiar chat completions format

Conclusion#

  • Combining OpenRouter with LLM CLI gives you a powerful, flexible way to experiment with and use LLMs from the terminal

    • Install once, configure your API key, and you have access to dozens of models
    • Switch between models with a single flag change
    • Use the direct API for scripting and automation

  • For more information: