Copilot Auto

Adds GitHub Copilot's Auto model to OpenCode V2. Pick github-copilot/auto and Copilot chooses which of your available models handles each prompt, the same way Auto works in VS Code.

Requires OpenCode 2.x. For OpenCode 1, use 0.1.x.

Setup

Connect GitHub Copilot in OpenCode (/connect), then add the plugin to opencode.jsonc:

{
  "$schema": "https://opencode.ai/config.json",
  "plugins": ["@opeoginni/opencode-copilot-auto"]
}

Restart OpenCode and select Auto under GitHub Copilot in the model picker. It only appears once Copilot is connected.

How it works

Each request to auto is sent to Copilot's routing endpoint, which picks a model from those available on your plan. The plugin then makes the real request with that model over its native protocol (Responses for GPT-5 and newer, Chat Completions otherwise), so tools, images and reasoning behave exactly as they do when you select the model directly.

Routing happens once per user prompt; tool calls within the same turn reuse the choice.

Options

{
  "plugins": [
    {
      "package": "@opeoginni/opencode-copilot-auto",
      "options": { "sticky": true, "notifications": true }
    }
  ]
}

Option	Default	Description
sticky	false	false: Copilot picks a model for every prompt. true: the first model Copilot picks is kept for the whole session.
notifications	false	Show a toast naming the model Copilot picked. Fires once per prompt, or once per session when sticky is on. Mostly useful while developing.

OpenCode does not record which model answered on the message itself, so the toast is the only place the choice is visible.

Development

bun install
bun run check
bun test
bun run build

To try the plugin locally without publishing, copy the example config, build, and start an isolated OpenCode in this directory:

cp opencode.example.jsonc opencode.jsonc
bun run build
opencode --standalone

opencode.jsonc points at ./dist, so rebuild after changes. Running with --standalone keeps the test server separate from your regular OpenCode service.