Claude Code Openrouter

OpenRouter Claude Code Agent

Use external LLMs (Kimi, Grok, Gemini, GLM, GPT-5) directly from Claude Code by mentioning them in your prompts.

Setup

New Project

Clone this repo and add your API key:

Get your API key from OpenRouter.ai

Existing Project

Run this from your project directory:

Then add your API key:

Usage

In Claude Code, just mention the model name:

- "Use kimi to write a blog post about AI"
- "Ask grok to explain quantum computing"
- "Have gemini create a marketing presentation and use @marketing/project-plan.md as context"

Add "external-llm" to your prompt if it doesn't trigger automatically.
First Run: Claude Code will ask for approval the first time. Select "Yes, and don't ask again..." to auto-approve future calls.

Available Models

Default models configured in models.conf:
- kimi - Moonshot AI's K2 model (great for creative writing)
- grok - Grok 4 from X.AI
- grok-fast - Grok 4 Fast from X.AI
- glm - GLM-4.6 from Zhipu AI
- gemini - Gemini 2.5 Pro from Google
- gpt-5 - GPT-5 from OpenAI

Adding Your Own Models

Edit models.conf to add any OpenRouter model:

Find model IDs at OpenRouter.ai/models

Output Locations

Default behavior - saves to llm-outputs/:

Custom paths - specify where to save:

Direct CLI Usage

You can also use the script directly from the command line:

CLI Options

- --model - Which model to use (from models.conf)
- --query - Your prompt
- --output - Where to save the response
- --system-prompt - Add instructions
- --context-file - Include file content
- --temperature - Creativity (0-2, default 0.7)
- --max-tokens - Max length (default 4000)

File Locations

- API key: .env or ~/.openrouter.env
- Model config: models.conf
- Generated files: llm-outputs/
- Agent config:…