GitHub Copilot

Copy Markdown View Source

GitHub Copilot exposes an OpenAI-compatible Chat Completions API at https://api.githubcopilot.com. ReqLLM supports it through the :github_copilot provider.

Configuration

Use a GitHub token that has access to Copilot:

export COPILOT_GITHUB_TOKEN=gho_...

By default, github_copilot_auth is :token. ReqLLM checks these token sources in order:

  1. The :api_key request option
  2. Application config config :req_llm, :github_copilot_api_key, token
  3. COPILOT_GITHUB_TOKEN
  4. GH_TOKEN
  5. GITHUB_TOKEN

If none is configured, ReqLLM returns an error. Running gh auth token requires an explicit opt-in:

ReqLLM.generate_text("github_copilot:gpt-4o-mini", "Hello!", github_copilot_auth: :auto)

Use :auto to check configured tokens first and then fall back to the GitHub CLI, or :gh to use only the GitHub CLI. Both modes execute the gh executable found on PATH.

You can also pass a token per request:

ReqLLM.generate_text("github_copilot:gpt-4o-mini", "Hello!", api_key: token)

Streaming

{:ok, response} =
  ReqLLM.stream_text(
    "github_copilot:gpt-4o-mini",
    "Reply with one sentence.",
    max_tokens: 120
  )

response
|> ReqLLM.StreamResponse.tokens()
|> Enum.each(&IO.write/1)

Provider Options

Provider options can be passed either top-level or nested under :provider_options.

ReqLLM.stream_text(
  "github_copilot:gpt-4o-mini",
  "Hello!",
  provider_options: [
    github_copilot_integration_id: "vscode-chat",
    github_copilot_auth: :token
  ]
)

github_copilot_integration_id

  • Type: String
  • Default: "vscode-chat"
  • Purpose: Sets the required Copilot-Integration-Id request header.

github_copilot_auth

  • Type: :auto | :token | :gh

  • Default: :token
  • Purpose: Controls credential lookup.

:token only uses configured token sources and returns an error if none is available. Explicitly set github_copilot_auth: :auto to check configured tokens first and then fall back to gh auth token, or github_copilot_auth: :gh to use only the GitHub CLI.

Model Specs

Copilot model availability is account-dependent. ReqLLM accepts unverified Copilot model IDs:

ReqLLM.model!("github_copilot:gpt-4o-mini")
ReqLLM.model!("github_copilot:claude-sonnet-4.5")

If you need to use a different Copilot-compatible host, set base_url:

model =
  ReqLLM.model!(%{
    provider: :github_copilot,
    id: "gpt-4o-mini",
    base_url: "https://api.individual.githubcopilot.com"
  })

ReqLLM.stream_text(model, "Hello!")

Token Exchange 404s

Some community examples call https://api.github.com/copilot_internal/v2/token before calling Copilot. That endpoint is internal and can return 404 Not Found for valid GitHub CLI OAuth tokens or account types.

ReqLLM does not depend on that exchange. It sends the resolved token directly to https://api.githubcopilot.com as a bearer token, matching the path used by the successful streaming request.