Skip to main content
Factory Droid CLI (command droid) is Factory AI’s terminal AI software-engineering agent.
Protocol: OpenAI-compatible
  • Base URL: https://api.orcarouter.ai/v1 (includes the /v1 suffix)

​
Install

macOS/Linux: 
curl -fsSL https://app.factory.ai/cli | sh
Windows (PowerShell): 
irm https://app.factory.ai/cli/windows | iex
The installed command is droid.

​
Configure

Config file: ~/.factory/config.json. 
{
  "custom_models": [
    {
      "model_display_name": "OrcaRouter Auto",
      "model": "orcarouter/auto",
      "base_url": "https://api.orcarouter.ai/v1",
      "api_key": "sk-orca-...",
      "provider": "generic-chat-completion-api",
      "max_tokens": 8192
    }
  ]
}
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

droid exec -m "orcarouter/auto" --auto low "Reply with exactly: OK"
Expected output: OK.

​
Notes

  • Custom models go in the custom_models array.
  • Use provider: "generic-chat-completion-api" for the OpenAI-compatible protocol.
  • The base_url must include /v1.
  • --auto <low|medium|high> controls autonomy (how much the agent may do unattended) — it is not the reasoning control. See below for thinking.
VS Code extension: Factory’s official VS Code extension shares the same ~/.factory/config.json as the CLI — the custom_models entry above applies to both, with no separate in-editor endpoint setting. Two caveats: Factory only fully tests and benchmarks against the official Anthropic/OpenAI APIs, and a known issue can route subagents back to Anthropic instead of your custom model.

​
Enable thinking / reasoning

Control reasoning effort with the -r / --reasoning-effort flag (separate from --auto): 
droid exec -m "anthropic/claude-sonnet-4.6" -r high --auto low "What is 17*23?"
Levels: low / medium / high (default is per-model — Anthropic models default off, GPT-5 defaults medium). You can also set it persistently in settings.json via reasoningEffort. Verified: -r high passes through to OrcaRouter. (Note: custom_models entries have no reasoning field — control it via -r or settings.json.)