copilot-cli-llm-proxy

GitHub Copilot CLI with Archestra LLM Proxy

This example shows how to run GitHub Copilot CLI through Archestra's OpenAI-compatible LLM Proxy with a virtual key.

The flow is:

sequenceDiagram
  participant CLI as GitHub Copilot CLI
  participant Proxy as Archestra LLM Proxy
  participant Provider as Model provider

  CLI->>Proxy: OpenAI-compatible request<br/>Authorization: Bearer virtual key
  Proxy->>Proxy: Resolve virtual key, policy, logs, cost attribution
  Proxy->>Provider: Provider request
  Provider-->>Proxy: Model response
  Proxy-->>CLI: OpenAI-compatible response

Loading

Prerequisites

• Archestra running locally:
  • frontend: http://localhost:3000
  • backend: http://localhost:9000
• At least one configured LLM provider API key in Archestra.
• Node.js 20+.
• GitHub Copilot CLI:

npm install -g @github/copilot

Create a Copilot env file

The setup script logs into local Archestra, creates a virtual key, and writes Copilot's custom-provider environment variables to .env.copilot.

npm run setup

Defaults:

Variable	Default
ARCHESTRA_UI_BASE_URL	http://localhost:3000
ARCHESTRA_API_BASE_URL	http://localhost:9000
ARCHESTRA_EMAIL	admin@example.com
ARCHESTRA_PASSWORD	password
ARCHESTRA_PROVIDER	first available provider key
ARCHESTRA_MODEL	provider key's best model, or first matching model

Override them if needed:

ARCHESTRA_PROVIDER=azure ARCHESTRA_MODEL=gpt-4.1 npm run setup

Run Copilot through Archestra

source .env.copilot

copilot -p "Reply with exactly: archestra-copilot-cli-ok"

Expected output includes:

archestra-copilot-cli-ok

Depending on your GitHub Copilot organization policy, Copilot may print a warning about MCP servers before the model response. That warning does not affect the LLM Proxy path.

For an interactive coding session:

source .env.copilot
copilot

Notes

• GitHub authentication is not required when Copilot CLI runs with a custom provider.
• COPILOT_PROVIDER_TYPE is set to openai because Copilot speaks the OpenAI-compatible protocol. COPILOT_PROVIDER_BASE_URL still uses the selected Archestra provider path, for example /v1/azure/<llm-proxy-id> for an Azure-mapped virtual key.
• The virtual key value is written once to .env.copilot. Treat it like a secret.
• Delete or rotate the virtual key in Archestra after the demo if you no longer need it.