Skip to main content
Crush is Charm’s open-source terminal AI coding agent.
Protocol: OpenAI-compatible
  • Base URL: https://api.orcarouter.ai/v1 (includes the /v1 suffix)

​
Install

npm install -g @charmland/crush
The npm package is a launcher; the first run downloads the platform binary automatically (requires network access).

​
Configure

Config file: ~/.config/crush/crush.json (Windows: %USERPROFILE%\.config\crush\crush.json). 
{
  "$schema": "https://charm.land/crush.json",
  "providers": {
    "orcarouter": {
      "type": "openai",
      "base_url": "https://api.orcarouter.ai/v1",
      "api_key": "sk-orca-...",
      "models": [
        { "id": "orcarouter/auto", "name": "OrcaRouter Auto", "context_window": 200000, "default_max_tokens": 4096 }
      ]
    }
  }
}
Swap orcarouter/auto for a specific model if you prefer, e.g. anthropic/claude-opus-4.8 or google/gemini-2.5-flash. Model IDs always use the vendor/model format.

​
Run

crush run "Reply with exactly: OK"
Expected output: OK.

​
Notes

  • The base_url must include /v1.
  • First run downloads the binary, so initial startup needs network access.
  • models[].id uses the vendor/model format (e.g. orcarouter/auto, anthropic/claude-opus-4.8).

​
Enable thinking / reasoning

Set it per model in crush.json:
  • OpenAI-style models — "reasoning_effort": "high" (values low / medium / high), and mark the model "can_reason": true.
  • Anthropic models — "think": true (boolean toggle for extended thinking).
"models": [
  { "id": "anthropic/claude-sonnet-4.6", "name": "Sonnet 4.6", "context_window": 200000, "default_max_tokens": 4096, "can_reason": true, "reasoning_effort": "high" }
]
Verified: reasoning_effort: "high" works against OrcaRouter.