Back to guides

Crush Integration

Connect Crush to LLM Gateway, discover models, and verify a coding task.

Crush is a terminal coding agent. Connect its OpenAI-compatible provider to LLM Gateway with a pay-as-you-go or DevPass key.

Video walkthrough

Install

1pnpm add -g @charmland/crush2crush --version

Other installation options are in the Crush documentation.

Connect LLM Gateway

Create an API key in your dashboard, then export it in the terminal where you run Crush:

1export LLMGATEWAY_API_KEY="your_api_key"

Add this provider to your project's crush.json, merging it with existing settings:

1{2  "providers": {3    "llmgateway": {4      "name": "LLM Gateway",5      "type": "openai",6      "base_url": "https://api.llmgateway.io/v1",7      "api_key": "$LLMGATEWAY_API_KEY"8    }9  }10}

The key stays in your environment. Crush discovers the available models from the gateway when no explicit model list is configured.

Choose a model and code

1cd your-project2crush

Select LLM Gateway and a model with tool support from the live catalogue. With DevPass, choose a canonical model ID supported by your plan, without an upstream provider prefix.

Start with a small task that has a clear test. Review tool permission requests, inspect the changes, and run the tests. Check usage in the workspace that issued your API key.

Troubleshooting

  • Authentication failed: Check that LLMGATEWAY_API_KEY is exported in the same terminal and the key is active.
  • Models missing: Restart Crush after editing the provider configuration. Check the endpoint and key before adding a manual model list.
  • Tool calls fail: Confirm the selected model supports tools.
  • Provider unavailable: Gateway fallback can route requests to another eligible provider. See the routing documentation for pinning and fallback controls.

See the Crush configuration reference for permissions and additional settings.