Skip to main content
GitHub Copilot CLI is GitHub’s terminal coding assistant. It supports BYOK (bring your own key), which lets you point it at OrcaRouter.
Protocol: Anthropic (via COPILOT_PROVIDER_TYPE=anthropic)
  • Base URL: https://api.orcarouter.ai — without /v1
  • With /v1 the request path becomes /v1/v1/messages and fails with a 404

​
Install

npm install -g @github/copilot
copilot --version
Requires Node.js 22+.

​
Configure

Set these environment variables: 
export COPILOT_PROVIDER_TYPE=anthropic
export COPILOT_PROVIDER_BASE_URL="https://api.orcarouter.ai"   # NOTE: no /v1
export COPILOT_PROVIDER_API_KEY="sk-orca-..."
export COPILOT_MODEL="orcarouter/auto"                         # or e.g. anthropic/claude-opus-4.8

​
Run

copilot -p "Reply with exactly: OK" --allow-all
You should see OK. Drop -p for an interactive session.

​
Notes

  • Base URL must not include /v1 — with /v1 the request path becomes /v1/v1/messages and fails with a 404.
  • Non-interactive mode requires --allow-all (or set COPILOT_ALLOW_ALL).
  • Model names use the vendor/model format; orcarouter/auto is the simplest choice.
Use the CLI, not the VS Code extension. GitHub Copilot’s VS Code extension does not currently expose a custom OpenAI-compatible endpoint — its Add Models picker (Anthropic / OpenAI / Google / OpenRouter / Ollama / Azure) only lets you enter an API key, with no base-URL field, so it can’t be pointed at OrcaRouter. Connect through the Copilot CLI with the environment variables above instead.

​
Enable thinking / reasoning

The --effort flag does NOT work with BYOK custom models. Copilot CLI checks the model name against a built-in allow-list of reasoning-capable models; a custom vendor/model name isn’t on it, so --effort high is rejected with:
Error: Model "anthropic/claude-sonnet-4.6" does not support reasoning effort configuration (requested: "high").
When pointed at OrcaRouter, run without --effort — thinking is then governed by the model’s / upstream’s default behavior. (Verified.)