For the complete documentation index, see llms.txt. Markdown versions of all docs pages are available by appending .md to any docs URL.

Route agentgateway LLM traffic to models available through GitHub Copilot.

Configure GitHub Copilot as an LLM provider in agentgateway.

Before you begin

Have a GitHub Copilot credential that the Copilot API accepts, for an account with access to the models that you want to use. Agentgateway does not obtain, exchange, or refresh the credential. An arbitrary GitHub personal access token might not work.

By default, agentgateway finds the credential on its own, from an environment variable or from the files that the GitHub Copilot tools and the GitHub CLI write. For the lookup order, see GitHub Copilot backend authentication. To supply the credential directly, set params.apiKey.

Configuration

# yaml-language-server: $schema=https://agentgateway.dev/schema/config

llm:
  models:
  - name: "*"
    provider: copilot

To pass the credential from an environment variable instead of discovering it, set params.apiKey.

llm:
  models:
  - name: "*"
    provider: copilot
    params:
      apiKey: "$GH_COPILOT_TOKEN"

The provider also sets the protocol headers that the Copilot API expects, such as editor-version and x-initiator. For the full list, see GitHub Copilot backend authentication. To configure Copilot on a route instead of in llm.models, use the backendAuth: copilot method that the same page describes.

Example request

After you run agentgateway with the configuration from the previous section, send a Chat Completions request to the /v1/chat/completions endpoint. Use a model that your Copilot account can use.

curl -X POST http://localhost:4000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4.1",
    "messages": [{"role": "user", "content": "Hello from GitHub Copilot!"}]
  }'

Request formats

Copilot models do not all accept the same API format. Agentgateway chooses the formats that it sends to Copilot from the model name, and it converts the client request when the client calls a different format.

Tags in the model catalog take precedence over these built-in rules.

For example, a Responses request to gpt-4.1 is converted to Chat Completions, and a Messages request to a claude-* model passes through. A Chat Completions request to a model that accepts only Responses, such as gpt-5, fails.

Troubleshoot