Copilot Messages API Provider for Opencode

This plugin provides access to the api.githubcopilot.com/v1/messages for accessing Claude 4.x models via your Copilot subscription.

The built-in Copilot client in Opencode has first-class support for OpenAI's /responses endpoint, used by GPT-5.x models, but Anthropic (and Google) models fall back to the more limited /chat/completions endpoint.

If you use them side-by-side (which you can with this plugin), the performance degradation will be immediately noticeable. Copilot's /messages endpoint proxies the same endpoint used by the first-party Anthropic provider.

Setup

1. Add to the plugin array in opencode.json or opencode.jsonc:

  "plugin": [
    "opencode-copilot-messages@latest"
  ]

2. Run opencode auth login
3. Search or scroll to "other"
4. Enter "copilot-messages"
5. Finish OAuth flow in browser
6. Launch opencode

You will now see a new "copilot-messages" provider populated with all available models that support /responses, obtained from Copilot's /models endpoint.

Config

The plugin will work without any additional config, but for the best experience it is recommended to override the default limit.output and limit.context:

{
  "provider": {
    "copilot-messages": {
      "models": {
        "claude-opus-4.6": {
          "limit": {
            "context": 200000,
            "output": 64000
          }
        }
      }
    }
  }
}

Quota

Currently, the "Premium Request" cost per model is:

• Haiku: 0.33x
• Sonnet: 1x
• Opus 4.5: 3x
• Opus 4.6: 3x
• Opus 4.1: 10x

This plugin properly handles the "user-initiated" vs. "agent-initiated" headers so you only spend requests on actual prompts. Tool loops and subagents are agent-initiated.