Lab 0 — Configure GitHub Copilot CLI

Install @github/copilot via the official CLI instructions. Run copilot --version and copilot help providers.

To use a Foundry GPT Responses deployment, create an isolated COPILOT_HOME, copy this lab’s provider.env.example to a private temporary file outside the repository, replace the resource host and deployment name, and source it. Set COPILOT_PROVIDER_MODEL_ID to a well-known base model ID supported by your CLI and COPILOT_PROVIDER_WIRE_MODEL to your Foundry deployment name. Set COPILOT_MODEL to the configured model selection, and COPILOT_PROVIDER_AZURE_API_VERSION to the version required by your endpoint. Provide COPILOT_PROVIDER_API_KEY securely. An Entra bearer token is not part of the documented Azure BYOK recipe; only attempt it after independently validating your endpoint and CLI version. Set only one credential source. BYOK does not require GitHub sign-in. Do not commit keys/tokens or capture them in debug logs.

Remote Foundry deployment compatibility is not established merely by setting these variables: test authentication, streaming and tool calling on your deployment and record any unsupported combinations.

If you instead choose the hosted Copilot path, do not source the BYOK variables; run copilot login and select a model your account can use. This is not a Foundry run and must be labeled accordingly.

In a disposable directory run copilot -C "$workdir" -i "Say hello in one sentence." (create it first with workdir="$(mktemp -d)"). Inspect the provider and model selected. The CLI may expose built-in tools even for this prompt; do not call it a bare model request. Remove temporary workspace, config and COPILOT_HOME after the exercise and unset credential variables. See provider reference and copilot help providers for version-specific provider behavior.